A new energy station frequency control parameter collaborative setting method and application thereof

Abstract

The application discloses a new energy station frequency control parameter collaborative setting method and application thereof, and the method comprises the following steps: constructing an optimized parameter vector x containing a network type and a network type station heterogeneous control parameter; establishing a comprehensive objective function f(x) which fuses the system frequency deviation minimization and the parameter adjustment cost minimization, and generating an initial training sample database corresponding to the comprehensive objective function based on sampling; constructing an anisotropic Kriging surrogate model based on the initial training sample database; constructing a sparse degree weighted adaptive expected improvement SW-AEI adding point criterion based on the anisotropic Kriging surrogate model, and screening an optimal candidate vector with the maximum sparse degree weighted adaptive expected improvement value; updating a new sample to the training sample database to recalibrate the anisotropic sensitivity hyperparameters in the anisotropic Kriging surrogate model until a convergence condition is met, and outputting the optimal collaborative control parameter.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to a method for coordinated tuning of frequency control parameters at new energy power plants and its application. Background Technology

[0002] Currently, renewable energy sources, primarily wind and solar power, are seeing a continuous increase in penetration into my country's new power system, gradually establishing their status as core power sources. However, unlike traditional synchronous generators with physical rotating mass, renewable energy power plants typically connect to the grid via power electronic converters. Under conventional control strategies, these interface devices struggle to proactively respond to system frequency changes and cannot provide the necessary inertial support. Especially in scenarios with a high proportion of renewable energy generation from wind and solar power, the system's equivalent inertia and primary frequency regulation capability will significantly decrease. Once active power disturbances occur, the system's rate of frequency change (RoCoF) will increase sharply, the frequency drop will deepen, and the risk of frequency exceeding limits can easily be triggered, threatening grid security.

[0003] Faced with this challenge, grid-following (GFL) and grid-forming (GFM) control technologies have become key to solving frequency problems. However, these two types of power plants differ fundamentally in their physical mechanisms: GFLs exhibit controlled current source characteristics, primarily following the grid frequency by simulating droop characteristics; while GFMs exhibit controlled voltage source characteristics, actively constructing voltage and frequency, providing instantaneous inertia support similar to a synchronous machine. In current engineering practice, the inertia time constant and frequency regulation parameters of renewable energy power plants are mostly preset to fixed values ​​by manufacturers. This static configuration cannot perceive the time-varying inertia requirements of the grid under different time periods and different unit combinations, and it also ignores the different roles played by GFLs and GFMs in frequency response. If a unified or independent tuning strategy is adopted for both, it is very easy to lead to unreasonable parameter configuration, causing dynamic conflicts between control loops, and even causing system oscillations.

[0004] Furthermore, existing parameter optimization methods face two major bottlenecks: First, the contradiction between computational efficiency and accuracy; high-fidelity electromagnetic transient simulations are extremely time-consuming, making it difficult to meet online tuning requirements; and traditional surrogate models (such as the standard Kriging model and response surface methodology) typically assume that all input parameters have isotropic spatial correlations, failing to effectively identify and handle the significant sensitivity differences between GFL and GFM heterogeneous parameters to the system's frequency response, leading to a significant decrease in model prediction accuracy in the parameter coupling region. Second, the limitations of optimization strategies; high-dimensional nonlinear parameter spaces contain numerous local optimum traps, and conventional intelligent optimization algorithms or simple expectation improvement strategies often struggle to achieve a dynamic balance between exploration and development, easily getting trapped in local optima or engaging in ineffective searches in flat regions, failing to quickly converge to the globally optimal cooperative parameter combination.

[0005] Therefore, overcoming the modeling difficulties caused by the sensitivity differences of heterogeneous parameters, establishing an efficient collaborative tuning mechanism that balances computational efficiency and global optimization capabilities, and dynamically optimizing virtual inertia and primary frequency regulation parameters based on the real-time operating conditions of the system have become urgent technical challenges to overcome in improving the frequency security level of new power systems. Summary of the Invention

[0006] In view of this, the present invention provides a method for collaborative tuning of frequency control parameters of new energy power stations and its application, which can at least solve the problems of modeling difficulties and low optimization efficiency caused by the sensitivity differences of heterogeneous parameters in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for coordinated tuning of frequency control parameters for new energy power stations includes the following steps:

[0009] S1: Based on the real-time operating status of the power system and the configuration information of new energy power plants, construct a parameter vector to be optimized, which includes heterogeneous control parameters of grid-connected and grid-connected power plants. Establish a comprehensive objective function that minimizes both the frequency deviation and parameter adjustment costs of the fusion system. And generated based on sampling The corresponding initial training sample database;

[0010] S2: Construct an anisotropic Kriging surrogate model based on the initial training sample database;

[0011] S3: Improving the SW-AEI point addition criterion by constructing a sparsity-weighted adaptive expectation criterion based on the anisotropic Kriging surrogate model. By adjusting The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. ;

[0012] S4: Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update the training sample database to recalibrate the anisotropic sensitivity hyperparameters in the anisotropic Kriging surrogate model. The process continues until the convergence condition is met, at which point the optimal cooperative control parameters are output.

[0013] Preferably, the parameter vector to be optimized in S1 ,in and These are the virtual inertia and sag coefficient of the grid-type station, respectively. The virtual inertia and sag coefficient of the grid-type station.

[0014] Preferably, in S1, the comprehensive objective function Its constraints are:

[0015] ;

[0016] ;

[0017] in, The vector of parameters to be optimized; and These are the frequency performance index weights and parameter adjustment cost weights, respectively. This refers to the simulation evaluation duration; For the time variable in the simulation process; For time The deviation of the system's inertial center frequency at that location; The normalization factor for the benchmark value; The index of an element in the parameter vector. For parameter vectors The Middle The adjustment penalty factor for each control parameter is used to distinguish the difference in adjustment costs between grid-connected and grid-constructed units; The parameter vector to be optimized The first in The current values ​​of each parameter component; , and They are respectively as follows: The initial value, physical upper limit value, and physical lower limit value; and These are the current parameter vectors to be optimized. The maximum rate of frequency change and the lowest frequency point of the lower system; The maximum allowable rate of change of frequency is the safety threshold. This is the lower limit threshold for system frequency safety. and These are the lower bound vector and upper bound vector of the parameter vector to be optimized, respectively.

[0018] Preferably, based on sampling generation The specific contents of the corresponding initial training sample database include: generating an initial sample set in the parameter feasible region using optimal Latin hypercube sampling, obtaining the corresponding objective function value through high-fidelity time-domain simulation, and constructing the initial training sample database.

[0019] Preferably, in S2, the anisotropic Kriging surrogate model uses an anisotropic Gaussian kernel function as the covariance function and utilizes maximum likelihood estimation to automatically identify and quantify the differential sensitivity weights of the control parameters of the following network type and the network structure type on the frequency response of the system, thereby achieving the decoupling and characterization of heterogeneous physical characteristics at the mathematical level and establishing a parameter-response mapping relationship.

[0020] The anisotropic Kriging proxy model is as follows:

[0021] ;

[0022] in, To predict the response value; This represents the mean of the global trend term. With a mean of 0 and a variance of Gaussian random processes, covariance function of Gaussian random processes Anisotropic Gaussian kernel function is used:

[0023] ;

[0024] in, For parameter dimensions; For anisotropic sensitivity hyperparameters, i.e., corresponding to the first... The characteristic length scale hyperparameter of the dimensional parameter component is obtained through maximum likelihood estimation optimization, and its numerical magnitude characterizes the dimensional parameter component. The sensitivity weights of the dimensional parameter components to the system frequency response are used to achieve automatic identification of heterogeneous parameter characteristics between the following network type and the structured network type. and These are the first and second samples in the training sample set, respectively. The and the first A vector of sample parameters; and Sample parameter vectors and In the The parameter component values ​​of the dimension; For smoothing parameters.

[0025] Preferably, a spatial sparsity potential function is introduced in S3 to construct a sparsity-weighted adaptive expectation improved point addition criterion. The specific content is as follows:

[0026] ;

[0027] in, This is a standard expected improvement item; Let be the spatial sparsity potential function; It is an adaptive dynamic balance factor; This is the current training sample database.

[0028] Preferred, standard expected improvement items for:

[0029] ;

[0030] in, The minimum value of the objective function in the current training sample database; and For the anisotropic Kriging surrogate model, the candidate vectors are respectively... The predicted response value and the root mean square error of the prediction; and These are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0031] Preferred spatial sparsity potential function for:

[0032] ;

[0033] in, Represents candidate vectors To the current training sample database The minimum Euclidean distance between all sample vectors in the dataset; This is the distance attenuation coefficient.

[0034] Preferred, adaptive dynamic equilibrium factor for:

[0035] ;

[0036] in, This represents the current iteration number; These are the initial balance coefficients; is the decay rate constant.

[0037] Preferably, in S3, by adjusting The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. The method is as follows:

[0038] Within the parameter design space, adjust The adaptive dynamic balance factor is determined, and the corresponding sparseness-weighted adaptive expected improvement value is obtained. This allows us to select the vector that maximizes the sparseness-weighted adaptive expected improvement value and use it as the optimal candidate vector. :

[0039] .

[0040] Preferably, the specific content of S4 includes:

[0041] Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update to the training sample database, i.e.:

[0042] ;

[0043] in, This is the optimal candidate vector. The true objective function value obtained by calculating the optimal candidate vector using a power system simulation model;

[0044] New sample Update the training sample database and recalibrate the anisotropic sensitivity hyperparameters of the surrogate model based on incremental data. ;

[0045] Determine if the convergence condition is met. If it is, output the optimal cooperative control parameters; otherwise, return to step S3 to filter the optimal candidate vector again. .

[0046] A terminal device includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the above-mentioned method for coordinated tuning of frequency control parameters of a new energy power station.

[0047] A computer-readable storage medium storing a computer program / instructions thereon; when the computer program / instructions are executed by a processor, they implement the above-mentioned method for coordinated tuning of frequency control parameters of a new energy power station.

[0048] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method for coordinated tuning of frequency control parameters of new energy power stations and its application, which has the following beneficial effects:

[0049] 1. This invention achieves data decoupling and precise coordination of the physical characteristics of heterogeneous power stations, resolving the problem of dynamic characteristic conflicts. Addressing the engineering challenges of significant differences in physical characteristics between grid-connected and grid-connected power stations in new power systems, and the susceptibility to dynamic conflicts arising from traditional independent tuning, this invention constructs an anisotropic parameter correlation model. This model can automatically identify and quantify the differentiated sensitivity of parameters in different control modes to system frequency. Through collaborative optimization, the voltage source support characteristics of grid-connected power stations and the current source following characteristics of grid-connected power stations are made complementary in terms of parameters over time, significantly improving the system's frequency disturbance immunity under extreme operating conditions.

[0050] 2. This invention overcomes the efficiency bottleneck of online tuning, balancing global optimization with computational efficiency. The proposed active learning strategy based on Sparse Weighted Adaptive Expectation Improvement (SW-AEI) overcomes the shortcomings of traditional intelligent algorithms, which rely on massive simulation data and have long computation times. Through the dynamic nonlinear coupling of spatial sparsity potential and statistical improvement expectation, the algorithm possesses global exploration capabilities in the early stages of iteration, avoiding getting trapped in local optima, and can quickly converge to the exact solution in the later stages. This enables the method to adapt to real-time changes in power grid operation modes, providing dispatchers with fast and reliable parameter tuning suggestions, and effectively ensuring the frequency security of high-proportion renewable energy power grids. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating a method for coordinated tuning of frequency control parameters at new energy power stations provided by the present invention;

[0053] Figure 2 This is a schematic diagram of the collaborative tuning algorithm framework for frequency control parameters of new energy power stations based on the improved Kriging proxy model in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of the improved IEEE-39 node test system topology used in an embodiment of the present invention;

[0055] Figure 4 This is a comparison chart of the system frequency response curves before and after optimization under typical disturbance conditions in an embodiment of the present invention;

[0056] Figure 5 This is a bar chart comparing the maximum frequency deviation of the system before and after optimization under typical disturbance conditions according to an embodiment of the present invention;

[0057] Figure 6 This is a bar chart comparing the maximum frequency change rate of the system before and after optimization under typical disturbance conditions according to an embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] This invention provides a method for coordinated tuning of frequency control parameters for new energy power stations, such as... Figure 1 As shown, this method utilizes the adaptive sensing capability of the anisotropic Kriging model for heterogeneous parameter sensitivity, combined with a sparsity-weighted adaptive expectation improvement strategy, to achieve rapid and accurate coordination of control parameters under complex operating conditions. The method includes the following steps:

[0060] S1: Based on the real-time operating status of the power system and the configuration information of new energy power plants, construct a parameter vector to be optimized, which includes heterogeneous control parameters of grid-connected and grid-connected power plants. To balance frequency response performance with equipment adjustment costs, a comprehensive objective function is established to minimize both the frequency deviation and parameter adjustment costs of the fusion system. And generated based on sampling The corresponding initial training sample database;

[0061] S2: Construct an anisotropic Kriging surrogate model based on the initial training sample database;

[0062] S3: Improving the SW-AEI point addition criterion by constructing a sparsity-weighted adaptive expectation criterion based on the anisotropic Kriging surrogate model. By adjusting The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. ;

[0063] S4: Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update the training sample database to recalibrate the anisotropic sensitivity hyperparameters in the anisotropic Kriging surrogate model. The process continues until the convergence condition is met, at which point the optimal cooperative control parameters are output.

[0064] To further implement the above technical solution, the parameter vector to be optimized in S1 ,in and These are the virtual inertia and sag coefficient of the grid-type station, respectively. The virtual inertia and sag coefficient of the grid-type station.

[0065] It should be noted that:

[0066] Based on the power grid wide-area measurement system, real-time operational status data of the power system is acquired, mainly including the system network topology, the operating status and output distribution of each synchronous generator unit, and the overall network load level data. Simultaneously, configuration information of renewable energy power stations connected to the system is acquired, identifying the geographical distribution, capacity share, and interface control modes of grid-connected (GFL) and grid-forming (GFM) power stations. Based on this information, the control parameters to be optimized are identified, and a heterogeneous parameter vector to be optimized is defined. .

[0067] To improve the numerical stability of subsequent surrogate model training and eliminate parameter dimension differences, the parameter vectors were modified. Normalize it and map it to the unit hypercube space. Subsequent calculations were all performed in the normalized space, and only during the simulation verification stage was the value denormalized to physical values.

[0068] To further implement the above technical solution, in S1, the comprehensive objective function is... Its constraints are:

[0069] ;

[0070] ;

[0071] in, The vector of parameters to be optimized; and These are the frequency performance index weights and parameter adjustment cost weights, respectively. This refers to the simulation evaluation duration; For the time variable in the simulation process; For time The deviation of the system's inertial center frequency at that location; The normalization factor for the benchmark value; The index of an element in the parameter vector. For parameter vectors The Middle The adjustment penalty factor for each control parameter is used to distinguish the difference in adjustment costs between grid-connected and grid-constructed units; The parameter vector to be optimized The first in The current values ​​of each parameter component; , and They are respectively as follows: The initial value, physical upper limit value, and physical lower limit value; and These are the current parameter vectors to be optimized. The maximum rate of frequency change and the lowest frequency point of the lower system; The maximum allowable rate of change of frequency is the safety threshold. This is the lower limit threshold for system frequency safety. and These are the lower bound vector and upper bound vector of the parameter vector to be optimized, respectively.

[0072] It should be noted that:

[0073] Different values ​​are set for network-connected and grid-linked units. (For example This is to reflect the differences in adjustment costs.

[0074] To further implement the above technical solution, based on sampling generation The specific contents of the corresponding initial training sample database include: parameters to be optimized feasible domain Within this framework, an initial sample set is generated within the parameter feasible region using optimal Latin hypercube sampling, containing... A sample vector is generated to ensure the uniform distribution and space-filling of the initial samples in the parameter space. Each generated initial sample is substituted into a high-fidelity simulation model of the power system (such as an electromagnetic transient model) for parallel computation, and the objective function value corresponding to each sample is extracted. Construct the initial training sample database :

[0075] ;

[0076] To further implement the above technical solution, the anisotropic Kriging surrogate model in S2 adopts anisotropic Gaussian kernel function as covariance function and uses maximum likelihood estimation to automatically identify and quantify the differential sensitivity weights of the control parameters of the following network type and the network structure type on the frequency response of the system, thereby achieving the decoupling and characterization of heterogeneous physical characteristics at the mathematical level and establishing parameter-response mapping relationship.

[0077] The anisotropic Kriging proxy model is as follows:

[0078] ;

[0079] in, To predict the response value; This represents the mean of the global trend term. With a mean of 0 and a variance of Gaussian random processes are defined to address the significant differences in frequency response sensitivity between the parameters of the following network and the constructed network. covariance function For anisotropic Gaussian kernel functions:

[0080] ;

[0081] in, For parameter dimensions; For anisotropic sensitivity hyperparameters, i.e., corresponding to the first... The characteristic length scale hyperparameter of the dimensional parameter component is obtained through maximum likelihood estimation optimization, and its numerical magnitude characterizes the dimensional parameter component. The sensitivity weights of the dimensional parameter components to the system frequency response are used to achieve automatic identification of heterogeneous parameter characteristics between the following network type and the structured network type. and These are the first and second samples in the training sample set, respectively. The and the first A vector of sample parameters; and Sample parameter vectors and In the The parameter component values ​​of the dimension; For smoothing parameters.

[0082] It should be noted that:

[0083] In this embodiment, parameter dimensions Set to 4; smoothing parameter Take 2.

[0084] In addition, in this embodiment, a log-likelihood function is constructed to train the model parameters. Utilizing sample set information, the log-likelihood function... Represented as:

[0085] ;

[0086] in, The number of samples; This is the sample correlation matrix, where the elements are calculated by the aforementioned anisotropic Gaussian kernel function, and are used to characterize the degree of spatial correlation between sample points in the sample set. This is the variance estimate;

[0087] Optimize hyperparameters using maximum likelihood estimation (MLE) method By maximizing the log-likelihood function described above, the optimal correlation length for each dimension of the parameter is determined; larger values... This indicates that the corresponding control parameters have a significant impact on the system frequency response, thereby automatically identifying and quantifying the heterogeneous characteristics of GFL and GFM parameters.

[0088] Based on a trained anisotropic Kriging model, the formulas for the predicted response and prediction error are derived; for any unsampled vector Its predicted response value and root mean square error They are respectively:

[0089] ;

[0090] ;

[0091] in, This is the correlation vector between the predicted sample and the training sample.

[0092] To further implement the above technical solution and to balance global exploration and local development during the iterative optimization process, S3 introduces a spatial sparsity potential function to construct a sparsity-weighted adaptive expectation improvement point addition criterion. The specific content is as follows:

[0093] ;

[0094] in, This is a standard expected improvement item; Let be the spatial sparsity potential function; It is an adaptive dynamic balance factor; This is the current training sample database.

[0095] To further implement the above technical solutions, the standard expects improvements. for:

[0096] ;

[0097] in, The minimum value of the objective function in the current training sample database; and For the anisotropic Kriging surrogate model, the candidate vectors are respectively... The predicted response value and the root mean square error of the prediction; and These are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0098] To further implement the above technical solution, the spatial sparsity potential function... for:

[0099] ;

[0100] in, Represents candidate vectors To the current training sample database The minimum Euclidean distance between all sample sets in the dataset; This is the distance decay coefficient, used to adjust the decay rate of sparsity rewards.

[0101] To further implement the above technical solution, an adaptive dynamic balance factor is needed. for:

[0102] ;

[0103] in, This represents the current iteration number; These are the initial balance coefficients; The decay rate constant;

[0104] The adaptive dynamic balance factor formula allows the algorithm to focus on sparsity exploration in the early stages and on expected improvement and development in the later stages.

[0105] To further implement the above technical solution, S3 is adjusted... The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. The method is as follows:

[0106] Within the parameter design space, adjust The adaptive dynamic balance factor is determined, and the corresponding sparseness-weighted adaptive expected improvement value is obtained. This allows us to select the vector that maximizes the sparseness-weighted adaptive expected improvement value and use it as the optimal candidate vector. :

[0107] .

[0108] It should be noted that:

[0109] In this embodiment, a particle swarm optimization algorithm or a genetic algorithm is used to find the optimal parameter space. The vector that is maximized is used as the best candidate vector for the next round. .

[0110] To further implement the above technical solution, the specific content of S4 includes:

[0111] Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update to the training sample database, i.e.:

[0112] ;

[0113] in, This is the optimal candidate vector. The true objective function value obtained by calculating the optimal candidate set using a power system simulation model;

[0114] New sample Update the training sample database and recalibrate the anisotropic sensitivity hyperparameters of the surrogate model based on incremental data. ;

[0115] Determine if the convergence condition is met. If it is, output the optimal cooperative control parameters; otherwise, return to step S3 to filter the optimal candidate vector again. .

[0116] It should be noted that:

[0117] In this embodiment, the convergence condition is reaching the maximum allowed number of iterations. , or continuous The relative improvement rate of the secondary objective function value is lower than a preset threshold. .

[0118] If the convergence condition is not met, the maximum likelihood estimation in S3 is re-executed to update the anisotropic hyperparameters with incremental data, thereby correcting the model's perception of parameter sensitivity. Then, SW-AEI is executed to continue the iteration.

[0119] If the convergence condition is met, the iteration terminates. Output the current database. The parameter combination corresponding to the sample with the smallest objective function value. As the final coordinated tuning result, this parameter is sent to the controllers of each grid-connected and grid-structured power station to complete the online tuning.

[0120] like Figure 2 As shown, by employing the method of this embodiment of the invention, based on the real-time operating status data of the system, and utilizing the constructed anisotropic Kriging proxy model and the sparsity-weighted adaptive expectation improvement strategy, it is possible to quickly and accurately find the coordinated control parameters of grid-connected and grid-connected power stations that meet the system frequency security constraints and have the lowest adjustment cost. This solves the problem of difficult optimization of parameter coupling of heterogeneous power stations and effectively improves the frequency stability of the power system when dealing with power disturbances.

[0121] A terminal device includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the above-mentioned method for coordinated tuning of frequency control parameters of a new energy power station.

[0122] A computer-readable storage medium storing a computer program / instructions thereon; when the computer program / instructions are executed by a processor, they implement the above-mentioned method for coordinated tuning of frequency control parameters of a new energy power station.

[0123] The invention will be further illustrated below through simulation experiments:

[0124] An improved New England 39-bus (IEEE 39-bus) test system was built in a power system simulation platform (Matlab / Simulink). To simulate the characteristics of renewable energy plants with a high proportion of converters, the synchronous generator unit at node 30 in the original model was replaced with a 260MW grid-joined (GFL) renewable energy plant, and the synchronous generator unit at node 36 was replaced with a 220MW grid-connected (GFM) renewable energy plant. The remaining nodes (31, 32, 33, 34, 35, 37, 38, 39) still retain synchronous generator units. The network topology of the system is as follows: Figure 4 As shown.

[0125] In this simulation model, the control parameters to be optimized and their adjustable ranges are set as follows: the virtual inertial time constants of the grid-type and network-structured power stations. , The value range is set to 4-12 s, and the initial fixed value before optimization is set to 6 s; the primary frequency modulation droop coefficient , The values ​​are all set to a range of 10-50, and the initial fixed values ​​before optimization are all set to 20. The simulation environment is based on a workstation with an Intel Core i7 processor and 16GB of memory, and a fixed step size of 0.01s is used for solving the problem. The duration of a single simulation is set to 50s.

[0126] To verify the effectiveness of the proposed collaborative tuning method based on sparse weighted adaptive expectation improvement (SW-AEI), the algorithm is first initialized: an initial sample set is generated using optimal Latin hypercube sampling, and the number of samples is... The number is set to 150. This represents the number of candidate vectors selected in each iteration. The maximum number of iterations is set to 500, with 20 new high-potential sample vectors added each round. The frequency performance index is set to 20 times. In the comprehensive objective function, the frequency performance index has a weight. With parameter adjustment cost weight The initial values ​​were all set to 0.5 to give equal importance to the frequency stability of the system and the economy of equipment adjustment.

[0127] To fully illustrate the adaptability of the proposed method, this invention constructs four typical operating scenarios (C1-C4) to simulate the power grid's operation under different load periods and different unit combinations: Scenario C1 (Low Inertia, High Risk): Simulating a low-load period at night, two synchronous generators located at nodes 38 and 39 are taken offline, significantly reducing the system's total physical inertia and relatively increasing the penetration rate of new energy sources. Scenario C2 (Sudden Fault): Simulating daytime operation, units located at nodes 33 and 37 are momentarily disconnected from the grid due to a sudden fault, posing a threat of a sharp drop in inertia to the system. Scenario C3 (Weak Support): Simulating a specific maintenance method, where units located at nodes 31 and 34 are taken offline, weakening local voltage support. Scenario C4 (Benchmark Full Operation): All synchronous generators and new energy power plants in the system are put into operation as a reference. At the start of the simulation for each of the above scenarios, a sudden increase in active power load (330MW) is set, and the proposed algorithm is used to collaboratively optimize the control parameters of GFL and GFM power plants.

[0128] According to the steps described in this invention, the optimized frequency control parameters of the system under different operating conditions can be obtained, as shown in Table 1. The results show that the cooperative tuning method proposed in this invention exhibits significant adaptability under all operating conditions.

[0129] Table 1. Frequency control parameter results under various operating conditions (C1-C4)

[0130] ;

[0131] Figure 4 , Figure 5 and Figure 6 The inertial center frequency response curves and frequency indicators of the system under four operating conditions are presented, using fixed parameters and the optimized parameters of this invention.

[0132] from Figure 4 It can be seen that in the initial stage after the disturbance, the slope of the optimized frequency curve decreases significantly, indicating that the optimized virtual inertia parameters play a key role in inertial support and effectively suppress RoCoF. During the frequency recovery stage, the optimized curve exhibits smaller overshoot and can stabilize at the quasi-steady-state frequency more quickly, verifying the effective coordination between the fast primary frequency regulation parameters and the grid-based control of the grid-connected power station. Simulation results fully demonstrate that the algorithm based on the improved Kriging surrogate model possesses the ability to sense the system's inertia state and differences in frequency regulation resources, and can adaptively output optimal control parameters under different unit combinations.

[0133] In summary, the proposed method for collaborative tuning of frequency control parameters for renewable energy power plants, based on an improved Kriging proxy model, utilizes a sparsity-weighted adaptive expectation improvement strategy to achieve online dynamic optimization of control parameters for both grid-connected and grid-connected power plants, taking into account the current system unit configuration and operating status. This method not only improves frequency quality under baseline operating conditions with sufficient inertia but also, under extreme operating conditions with low inertia and high risk, fully explores and coordinates the frequency regulation potential of heterogeneous power plants, ensuring that system frequency indicators remain within safe operating boundaries and effectively mitigating the frequency stability crisis caused by a high proportion of renewable energy integration.

[0134] The present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0139] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for coordinated tuning of frequency control parameters for new energy power stations, characterized in that, Includes the following steps: S1: Based on the real-time operating status of the power system and the configuration information of new energy power plants, construct a parameter vector to be optimized, which includes heterogeneous control parameters of grid-connected and grid-connected power plants. Establish a comprehensive objective function that minimizes both the frequency deviation and parameter adjustment costs of the fusion system. And generated based on sampling The corresponding initial training sample database; S2: Construct an anisotropic Kriging surrogate model based on the initial training sample database; S3: Improving the SW-AEI point addition criterion by constructing a sparsity-weighted adaptive expectation criterion based on the anisotropic Kriging surrogate model. By adjusting The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. ; S4: Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update the training sample database to recalibrate the anisotropic sensitivity hyperparameters in the anisotropic Kriging surrogate model. Continue until the convergence condition is met, and then output the optimal cooperative control parameters. The parameter vector to be optimized in S1 ,in and These are the virtual inertia and sag coefficient of the grid-type station, respectively. The virtual inertia and droop coefficient of the grid-type station; In S1, the comprehensive objective function Its constraints are: ； ； in, and These are the frequency performance index weights and parameter adjustment cost weights, respectively. This refers to the simulation evaluation duration; For the time variable in the simulation process; For time The deviation of the system's inertial center frequency at that location; The normalization factor for the benchmark value; The index of an element in the parameter vector. For parameter vectors The Middle The adjustment penalty factor for each control parameter is used to distinguish the difference in adjustment costs between grid-connected and grid-constructed units; The parameter vector to be optimized The first in The current values ​​of each parameter component; , and They are respectively as follows: The initial value, physical upper limit value, and physical lower limit value; and These are the current parameter vectors to be optimized. The maximum rate of frequency change and the lowest frequency point of the lower system; The maximum allowable rate of change of frequency is the safety threshold. This is the lower limit threshold for system frequency safety. and These are the lower bound vector and upper bound vector of the parameter vector to be optimized, respectively.

2. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 1, characterized in that, Based on sampling generation The specific contents of the corresponding initial training sample database include: generating an initial sample set in the parameter feasible region using optimal Latin hypercube sampling, obtaining the corresponding objective function value through high-fidelity time-domain simulation, and constructing the initial training sample database.

3. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 1, characterized in that, In S2, the anisotropic Kriging surrogate model uses anisotropic Gaussian kernel function as covariance function and uses maximum likelihood estimation to automatically identify and quantify the differential sensitivity weights of the control parameters of the following network type and the network structure type on the frequency response of the system. This achieves the decoupling and characterization of heterogeneous physical characteristics at the mathematical level and establishes the parameter-response mapping relationship. The anisotropic Kriging proxy model is as follows: ； in, To predict the response value; This represents the mean of the global trend term. With a mean of 0 and a variance of Gaussian random processes, covariance function of Gaussian random processes Anisotropic Gaussian kernel function is used: ； in, For parameter dimensions; For anisotropic sensitivity hyperparameters, i.e., corresponding to the first... The characteristic length scale hyperparameter of the dimensional parameter component is obtained through maximum likelihood estimation optimization, and its numerical magnitude characterizes the dimensional parameter component. The sensitivity weights of the dimensional parameter components to the system frequency response are used to achieve automatic identification of heterogeneous parameter characteristics between the following network type and the structured network type. and These are the first and second samples in the training sample set, respectively. The and the first A vector of sample parameters; and Sample parameter vectors and In the The parameter component values ​​of the dimension; For smoothing parameters.

4. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 1, characterized in that, S3 introduces a spatial sparsity potential function to construct a sparsity-weighted adaptive expectation improved point addition criterion. The specific content is as follows: ； in, This is a standard expected improvement item; Let be the spatial sparsity potential function; An adaptive dynamic equilibrium factor; This is the current training sample database.

5. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 4, characterized in that, Standard Expected Improvements for: ； in, The minimum value of the objective function in the current training sample database; and For the anisotropic Kriging surrogate model, the candidate vectors are respectively... The predicted response value and the root mean square error of the prediction; and These are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

6. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 4, characterized in that, Spatial sparsity potential function for: ； in, Represents candidate vectors To the current training sample database The minimum Euclidean distance between all sample vectors in the dataset; This is the distance attenuation coefficient.

7. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 4, characterized in that, Adaptive dynamic balance factor for: ； in, This represents the current iteration number; These are the initial balance coefficients; This is the decay rate constant.

8. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 1, characterized in that, S3 is adjusted The adaptive dynamic balance factor in the selection process filters out the optimal candidate vector that maximizes the sparseness-weighted adaptive expected improvement value. The method is as follows: Within the parameter space to be optimized, adjust The adaptive dynamic balance factor is determined, and the corresponding sparseness-weighted adaptive expected improvement value is obtained. This allows us to select the vector that maximizes the sparseness-weighted adaptive expected improvement value and use it as the optimal candidate vector. : 。 9. The method for coordinated tuning of frequency control parameters for new energy power stations according to claim 1, characterized in that, The specific content of S4 includes: Will Substitute into the comprehensive objective function Verification calculations were performed to obtain the true objective function value. , will new sample Update to the training sample database, i.e.: ； in, This is the optimal candidate vector. The true objective function value obtained by calculating the optimal candidate vector using a power system simulation model; New sample Update the training sample database and recalibrate the anisotropic sensitivity hyperparameters of the surrogate model based on incremental samples. ; Determine if the convergence condition is met. If it is, output the optimal cooperative control parameters; otherwise, return to step S3 to filter the optimal candidate vector again. .

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory; characterized in that, The processor executes the computer program to implement the method for coordinated tuning of frequency control parameters for new energy power stations as described in any one of claims 1-9.

11. A computer-readable storage medium having a computer program / instructions stored thereon; characterized in that, When a computer program / instruction is executed by a processor, it implements the method for coordinated tuning of frequency control parameters for a new energy power station as described in any one of claims 1-9.

Images