Inertia configuration method and system based on power system fault whole process dynamic frequency safety boundary

Abstract

The invention discloses an inertia configuration method and system based on a power system fault whole process dynamic frequency safety boundary. A self-supervised learning encoder-decoder model is used to carry out alignment, deletion repair and noise elimination on multi-source time sequence data, and a unified high-quality measurement sequence is generated. A frequency dynamic estimation model fusing a factor graph and a physical constraint neural network is constructed, dynamic state quantities such as frequency deviation, change rate and power balance are obtained by combining factor graph reasoning and Sheng differential equation modeling, deep fusion of a physical model and data driving is achieved, and the whole-process evolvable frequency dynamic prediction capacity is obtained. And a whole-process frequency security boundary discrimination method fusing Neure-ODE and a monotonic neural network is provided, and intelligent discrimination of the security boundary is realized by using the monotonic neural network. And establishing a minimum inertia optimization model under the constraint of the safety boundary, and obtaining an optimal minimum inertia configuration scheme meeting the whole-process frequency safety boundary. And the safe and stable operation capability of the power system is improved.

Description

Technical Field

[0001] This invention relates to an inertia configuration method and system based on the dynamic frequency safety boundary of the entire fault process in a power system, and belongs to the field of power systems. Background Technology

[0002] With the continuous advancement of new power system construction, the high proportion of renewable energy grid connection has led to a decline in system inertia, making frequency stability issues increasingly prominent. Traditional frequency security assessment methods mostly rely on offline simulation and deterministic criteria, making it difficult to adapt to complex scenarios involving the dynamic characteristics of the entire fault process and the coupling of multiple time scales. In existing technologies, fault initiation identification often relies on manually set thresholds or a single data source, which is susceptible to noise and measurement deficiencies; frequency dynamic estimation typically employs simplified physical models or purely data-driven methods, lacking integration with the inherent mechanisms of system power balance, resulting in insufficient prediction accuracy; frequency security boundary discrimination is mostly based on static rules or empirical thresholds, failing to accurately reflect the dynamic security margin of the frequency trajectory throughout the fault process; and in terms of inertia allocation, existing methods often ignore the influence of spatial distribution characteristics and uncertain disturbances, making it difficult to achieve optimal and robust minimum inertia allocation. Therefore, there is an urgent need to develop a frequency dynamic boundary analysis method that can integrate multi-source data, embed physical mechanisms, and possess intelligent discrimination and optimization capabilities. Summary of the Invention

[0003] This invention provides an inertia configuration method and system based on the dynamic frequency safety boundary of the entire power system fault process, which solves the problems disclosed in the background art.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] Inertia configuration method based on dynamic frequency safety boundary throughout the entire power system fault process:

[0006] Acquire multi-source time-series data of the entire process of power system faults;

[0007] Multi-source time series data are preprocessed to generate a unified measurement sequence;

[0008] The measurement sequence is input to a pre-constructed fusion factor map and a PINN frequency dynamic estimation model, and outputs the dynamic frequency of the entire fault process.

[0009] Based on the frequency trajectory, key frequency safety indicators are extracted from the entire fault process to construct frequency safety boundaries.

[0010] Based on the frequency safety boundary constraints, a minimum inertia optimization model is constructed to output the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process.

[0011] Furthermore, the method for preprocessing multi-source time series data to generate a unified measurement sequence is as follows:

[0012] Encoder-decoder models using self-supervised learning for multi-source time series data Repair and alignment are performed to obtain a uniform sequence. The Bayesian change point detection method is used to identify the time when the fault occurs. ;

[0013] ;

[0014] ;

[0015] In the formula: M is a mask matrix used to mark missing or abnormal data points; This indicates that Hadamard multiplication is performed point by point; For neural network encoders, For neural network encoders, both parameters are... ; It represents the sequence likelihood probability at time t when a new change has just occurred, assuming "run length=0".

[0016] Furthermore, the method for outputting the dynamic frequency of the entire fault process by inputting the measurement sequence into the pre-constructed fusion factor map and the PINN frequency dynamic estimation model is as follows:

[0017] Unified sequence As an observation Multi-source state estimation is performed using factor graph inference or unscented Kalman filtering to obtain the frequency deviation. ROCoF, dynamic status of mechanical / electrical power deviation;

[0018] ;

[0019] ;

[0020] In the formula: This represents the power system state vector. This is the state transition function; For input control quantity; The observation function maps the state to the observations; , The weighted L2 norm is used, and Q and R are the covariance matrices of process noise and observation noise. The estimated optimal system state;

[0021] Then use PINN for training, requiring Simultaneously satisfying both data and physical residuals, Neural-ODE is used to treat it as a "continuous-time differential equation," which is then integrated using an ODE solver to obtain... ;

[0022] ;

[0023] In the formula: For the estimated frequency deviation; The rate of change of frequency; Given a physical frequency model, the parameters Includes system inertia H and damping D; u is the input power control signal; It is an unknown dynamic compensation term learned by a neural network, with parameters... .

[0024] Furthermore, based on the frequency trajectory, the method for extracting key frequency security indicators throughout the fault process and constructing frequency security boundaries is as follows:

[0025] Based on Neural-ODE output Calculate frequency safety indicators: instantaneous RoCoF after a fault, minimum frequency, steady-state frequency, and time required for frequency drop.

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] In the formula: RoCoF is the rate of change of frequency. This represents the lowest frequency during the fault process. For the final steady-state frequency, This refers to the time it takes for the frequency to reach its lowest point.

[0031] Define a frequency safety set S. If a frequency process satisfies the requirements of maximum ROCoF, minimum frequency, steady-state deviation, and minimum point time threshold, then the power system is considered to be operating safely.

[0032] ;

[0033] In the formula: This is the maximum permissible value for RoCoF. The minimum allowable frequency threshold, For the rated frequency, For the allowable deviation of steady-state frequency, The minimum point arrival time threshold;

[0034] Learning safety boundary functions using monotonic neural networks Output the score s, if The power system is considered to be operating safely; this is achieved through monotonicity constraints. This ensures the physical consistency that "the greater the inertia, the safer it is";

[0035] ;

[0036] ;

[0037] ;

[0038] In the formula, The input feature vector includes system parameters and frequency indices, and y represents the security classification result: 0 = secure, 1 = insecure. This represents the total inertia of the power system.

[0039] Furthermore, based on the frequency safety boundary constraints, a minimum inertia optimization model is constructed, and the method for outputting the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process is as follows:

[0040] Under the premise of ensuring frequency security constraints, construct a partial Blue bar optimization problem to minimize the total power system output. Monotonic neural networks are used to determine the constraints. Express;

[0041] ;

[0042] ;

[0043] In the formula: Let be the inertia of the i-th synchronous machine. The virtual inertia provided for the j-th inverter This is the droop factor for frequency modulation. , These are the virtual inertia and damping control gain of the inverter, respectively.

[0044] Using graph neural networks to learn power grid topology And parameters, to obtain the sensitivity of nodal power perturbation to frequency. This is used to guide the allocation of inertia at different nodes; and combined with the Bruker bar optimization method, it optimizes the distribution of power perturbations. Within the Wasserstein sphere, the power system falls within a safe set. The probability is no less than ;

[0045] ;

[0046] ;

[0047] ;

[0048] In the formula: B is the line susceptance matrix, and M and R are the sets of virtual inertia and droop parameters, respectively. Let n be the power perturbation at node n. The true distribution of the disturbance; This is an empirical distribution, obtained through observed samples. For confidence level parameters, For Wasserstein distance The set of distributions with radius.

[0049] A second aspect of the present invention provides an inertia configuration system based on a dynamic frequency safety boundary throughout the entire power system fault process, comprising:

[0050] The multi-source time-series data module is used to acquire multi-source time-series data of the entire process of power system faults;

[0051] The data repair module is used to preprocess multi-source time-series data to generate a unified measurement sequence;

[0052] The frequency dynamic prediction module is used to input the measurement sequence into a pre-constructed fusion factor map and the frequency dynamic estimation model of PINN, and output the dynamic frequency of the entire fault process.

[0053] The frequency security boundary module is used to extract key frequency security indicators throughout the fault process based on the frequency trajectory and construct the frequency security boundary.

[0054] The inertia configuration module is used to construct a minimum inertia optimization model based on frequency safety boundary constraints, and output the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process.

[0055] A third aspect of the present invention provides a computer-readable storage medium for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0056] A fourth aspect of the present invention provides a computing device, comprising:

[0057] One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.

[0058] The beneficial effects achieved by this invention are as follows:

[0059] This invention provides support for the safe and stable operation of power systems through a fault initiation time identification method based on self-supervised learning and Bayesian change point detection, a frequency dynamic estimation model that integrates factor graphs and PINN, a full-process frequency safety boundary discrimination method that integrates Neural-ODE and monotonic neural networks, and a full-process frequency safety boundary optimal inertia configuration method based on distributed bar and GNN agent. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0062] Example 1

[0063] like Figure 1 As shown, this embodiment provides an inertia configuration method based on the dynamic frequency safety boundary throughout the entire power system fault process, including the following steps:

[0064] Step 1: Propose a fault initiation time identification method based on self-supervised learning and Bayesian change point detection. Utilizing a self-supervised learning encoder-decoder model, alignment, missing data repair, and noise removal are performed on multi-source time-series data such as SCADA and PMU to generate a unified and high-quality measurement sequence. Simultaneously, based on the Bayesian change point detection method, the initiation time of the fault is accurately identified, providing reliable input for subsequent dynamic modeling and safety assessment.

[0065] Step 2: Construct a frequency dynamic estimation model integrating factor graphs and PINN. Using the repaired measurement data as observation input, and leveraging factor graphs to fuse the system dynamics model and multi-source observations, dynamic state variables such as frequency deviation, rate of change, and power balance are obtained. Based on this, a frequency dynamic model is constructed with a physical model as its backbone, modified by Physically Informed Neural Networks (PINN) and Neural Ordinary Differential Equations (Neural-ODE). This achieves a deep fusion of physical constraints and data-driven approaches, thereby obtaining a fully evolvable frequency dynamic prediction capability.

[0066] Step 3: Propose a full-process frequency safety boundary discrimination method integrating Neural-ODE and monotonic neural networks. Based on the frequency trajectory output, extract key frequency safety indicators (RoCoF, minimum frequency, steady-state frequency, and arrival time of the minimum frequency point) throughout the fault process and construct a frequency safety set. Utilize a monotonic neural network to achieve intelligent discrimination of the frequency safety boundary, and ensure the physical consistency of the discrimination model by constraining the monotonicity of the system's total inertia, thereby quickly determining whether the system has crossed the safety boundary throughout the fault process.

[0067] Step 4: Propose an optimal inertia configuration method for the entire frequency safety boundary based on distributed Bruker bars and GNN proxy. Based on the frequency safety boundary constraints, a minimum inertia optimization model is constructed, jointly considering decision variables such as synchronous machine inertia, inverter virtual inertia, and frequency regulation droop parameters. The relationship between grid topology and sensitivity is learned through Graph Neural Networks (GNNs) to optimize the spatial configuration of inertia. A distributed Bruker bar optimization framework is introduced to ensure that the system still possesses a high safety probability under uncertain disturbance distribution conditions. Finally, the optimal minimum inertia configuration scheme that satisfies the entire frequency safety boundary is obtained.

[0068] For step one, self-supervised learning (Encoder-Decoder network) is used to process multi-source time series data. Repair and alignment were performed to resolve differences between low-frequency SCADA data, high-frequency PMU data, and incomplete data, resulting in a unified sequence. As shown in formula (1). Simultaneously, the Bayesian change point detection method is used to identify the time when the fault occurs. That is, to find the point where the statistical distribution of the time series changes significantly, as shown in formula (2).

[0069] (1);

[0070] (2);

[0071] In the formula: M is a mask matrix used to mark missing or abnormal data points; This indicates that Hadamard multiplication is performed point by point; For neural network encoders, For neural network encoders, both parameters are... ; It represents the sequence likelihood probability at time t assuming "run length=0" (a new change has just occurred).

[0072] For step two, the resulting unified sequence As an observation Multi-source state estimation is performed using factor graph inference or unscented Kalman filtering (UKF) to obtain frequency deviations. Dynamic states such as ROCoF, mechanical / electrical power deviation, etc. are shown in formulas (3)-(4).

[0073] (3);

[0074] (4);

[0075] In the formula: The system state vector includes , , , ; For the state transition function (physical prediction model); For input control quantity (frequency modulation power command); The observation function maps the state to the observations; , The weighted L2 norm is used, and Q and R are the covariance matrices of process noise and observation noise. This represents the estimated optimal system state.

[0076] Then use PINN for training, requiring Simultaneously satisfying both data and physical residuals, as shown in equation (5), Neural-ODE is used to treat it as a "continuous-time differential equation," and the result is obtained by integrating with the ODE solver. .

[0077] (5);

[0078] In the formula: For the estimated frequency deviation; The rate of change of frequency; For a known physical frequency model (such as a second-order oscillation equation), the parameters... Includes system inertia H, damping D, etc.; u is the input (power control signal). It is an unknown dynamic compensation term learned by a neural network, with parameters... .

[0079] For step three, based on the Neural-ODE output Calculate the frequency safety indicators: instantaneous RoCoF after the fault, minimum frequency, steady-state frequency, and time required for frequency drop, as shown in formulas (6)-(9).

[0080] (6);

[0081] (7);

[0082] (8);

[0083] (9);

[0084] In the formula: RoCoF is the rate of change of frequency. This represents the lowest frequency during the fault process. For the final steady-state frequency, This refers to the time it takes for the frequency to reach its lowest point.

[0085] Define the frequency safety set S as shown in formula (10), that is, as long as the frequency process meets the threshold requirements such as maximum ROCoF, minimum frequency, steady-state deviation and minimum point time, the system is considered to be safe to operate.

[0086] (10);

[0087] In the formula: This is the maximum permissible value for RoCoF. The minimum allowable frequency threshold, The rated frequency is 50 / 60Hz. For the allowable deviation of steady-state frequency, This is the threshold time for reaching the lowest point.

[0088] Finally, a monotonic neural network is used to learn the safety boundary function. Output the score s, if The system is considered safe based on monotonicity constraints. To ensure the physical consistency of "the greater the inertia, the safer", as shown in formulas (11)-(13).

[0089] (11);

[0090] (12);

[0091] (13);

[0092] In the formula, The input feature vector (including system parameters and frequency indices) is y, and the security judgment result is y (0=safe, 1=unsafe). This represents the total inertia of the system.

[0093] For step four, we construct a distributed bar optimization problem, with the goal of minimizing the total system size while ensuring frequency safety constraints. The monotonic neural network from step three is used to determine the constraints. The expression is shown in formulas (14)-(15).

[0094] (14);

[0095] (15);

[0096] In the formula: Let be the inertia of the i-th synchronous machine. The virtual inertia provided for the j-th inverter This is the droop factor for frequency modulation. , These are the inverter's virtual inertia and damping control gain, respectively.

[0097] Using graph neural networks (GNNs) to learn power grid topology And parameters, to obtain the sensitivity of nodal power perturbation to frequency. This guides the allocation of inertia to different nodes. Combined with the distributed bar optimization method, it ensures that even the distribution of power perturbations... Uncertainty (within the Wasserstein sphere), the system falls within a safe set. The probability is not lower than As shown in formulas (16)-(18).

[0098] (16);

[0099] (17)

[0100] (18);

[0101] In the formula: B is the line susceptance matrix, and M and R are the sets of virtual inertia and droop parameters, respectively. Let n be the power perturbation at node n. This represents the true distribution of the disturbance. This is an empirical distribution (obtained from observed samples). For confidence level parameters, For Wasserstein distance The set of distributions with radius.

[0102] Example 2

[0103] Based on Example 1, this example provides an inertia configuration system based on the dynamic frequency safety boundary of the entire power system fault process, including:

[0104] The multi-source time-series data module is used to acquire multi-source time-series data of the entire process of power system faults;

[0105] The data repair module is used to preprocess multi-source time-series data to generate a unified measurement sequence;

[0106] The frequency dynamic prediction module is used to input the measurement sequence into a pre-constructed fusion factor map and the frequency dynamic estimation model of PINN, and output the dynamic frequency of the entire fault process.

[0107] The frequency security boundary module is used to extract key frequency security indicators throughout the fault process based on the frequency trajectory and construct the frequency security boundary.

[0108] The inertia configuration module is used to construct a minimum inertia optimization model based on frequency safety boundary constraints, and output the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process.

[0109] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0110] A computer-readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform an inertia configuration method based on a dynamic frequency safety boundary throughout a power system fault.

[0111] A computing device includes one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for executing an inertia configuration method based on a dynamic frequency safety boundary for the entire process of power system faults.

[0112] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0113] 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.

[0114] 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.

[0115] 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.

[0116] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. An inertia configuration method based on the dynamic frequency safety boundary throughout the entire fault process of a power system, characterized in that: Acquire multi-source time-series data of the entire process of power system faults; Multi-source time series data are preprocessed to generate a unified measurement sequence; The measurement sequence is input to a pre-constructed fusion factor map and a PINN frequency dynamic estimation model, and outputs the dynamic frequency of the entire fault process. Based on the frequency trajectory, key frequency safety indicators are extracted from the entire fault process to construct frequency safety boundaries. Based on the frequency safety boundary constraints, a minimum inertia optimization model is constructed to output the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process.

2. The inertia configuration method based on the dynamic frequency safety boundary of the entire power system fault process according to claim 1, characterized in that: The method for preprocessing multi-source time series data to generate a unified measurement sequence is as follows: Encoder-decoder models using self-supervised learning for multi-source time series data Repair and alignment are performed to obtain a uniform sequence. The Bayesian change point detection method is used to identify the time when the fault occurs. ; ； ； In the formula: M is a mask matrix used to mark missing or abnormal data points; This indicates that Hadamard multiplication is performed point by point; For neural network encoders, For neural network encoders, both parameters are... ; It represents the sequence likelihood probability at time t when a new change has just occurred, assuming "run length=0".

3. The inertia configuration method based on the dynamic frequency safety boundary of the entire power system fault process according to claim 2, characterized in that: The method for outputting the dynamic frequency of the entire fault process by inputting the measurement sequence into the pre-constructed fusion factor map and the PINN frequency dynamic estimation model is as follows: Unified sequence As an observation Multi-source state estimation is performed using factor graph inference or unscented Kalman filtering to obtain the frequency deviation. ROCoF, dynamic status of mechanical / electrical power deviation; ； ； In the formula: This represents the power system state vector. This is the state transition function; For input control quantity; The observation function maps the state to the observations; , The weighted L2 norm is used, and Q and R are the covariance matrices of process noise and observation noise. The estimated optimal system state; Then use PINN for training, requiring Simultaneously satisfying both data and physical residuals, Neural-ODE is used to treat it as a "continuous-time differential equation," which is then integrated using an ODE solver to obtain... ; ； In the formula: For the estimated frequency deviation; The rate of change of frequency; Given a physical frequency model, the parameters Includes system inertia H and damping D; u is the input power control signal; It is an unknown dynamic compensation term learned by a neural network, with parameters... .

4. The inertia configuration method based on the dynamic frequency safety boundary of the entire power system fault process according to claim 3, characterized in that: Based on the frequency trajectory, the method for extracting key frequency safety indicators throughout the entire fault process and constructing frequency safety boundaries is as follows: Based on Neural-ODE output Calculate frequency safety indicators: instantaneous RoCoF after a fault, minimum frequency, steady-state frequency, and time required for frequency drop. ； ； ； ； In the formula: RoCoF is the rate of change of frequency. This represents the lowest frequency during the fault process. For the final steady-state frequency, This refers to the time it takes for the frequency to reach its lowest point.

5. Define a frequency safety set S. If a frequency process satisfies the requirements of maximum ROCoF, minimum frequency, steady-state deviation, and minimum point time threshold, then the power system is considered to be operating safely. ； In the formula: This is the maximum permissible value for RoCoF. The minimum allowable frequency threshold, For the rated frequency, For the allowable deviation of steady-state frequency, The minimum point arrival time threshold; Learning safety boundary functions using monotonic neural networks Output the score s, if The power system is considered to be operating safely; this is achieved through monotonicity constraints. This ensures the physical consistency that "the greater the inertia, the safer it is"; ； ； ； In the formula, The input feature vector includes system parameters and frequency indices, and y represents the security classification result: 0 = secure, 1 = insecure. This represents the total inertia of the power system.

6. The inertia configuration method based on the dynamic frequency safety boundary of the entire power system fault process according to claim 4, characterized in that: Based on frequency safety boundary constraints, the method for constructing a minimum inertia optimization model and outputting the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process is as follows: Under the premise of ensuring frequency security constraints, construct a partial Blue bar optimization problem to minimize the total power system output. Monotonic neural networks are used to determine the constraints. Express; ； ； In the formula: Let be the inertia of the i-th synchronous machine. The virtual inertia provided for the j-th inverter This is the droop factor for frequency modulation. , These are the virtual inertia and damping control gain of the inverter, respectively. Using graph neural networks to learn power grid topology And parameters, to obtain the sensitivity of nodal power perturbation to frequency. This is used to guide the allocation of inertia to different nodes; By combining the sub-Bruker optimization method, the distribution of power perturbation is improved. Within the Wasserstein sphere, the power system falls within a safe set. The probability is no less than ; ； ； ； In the formula: B is the line susceptance matrix, and M and R are the sets of virtual inertia and droop parameters, respectively. Let n be the power perturbation at node n. The true distribution of the disturbance; This is an empirical distribution, obtained through observed samples. For confidence level parameters, For Wasserstein distance The set of distributions with radius.

7. An inertia configuration system based on dynamic frequency safety boundaries throughout the entire power system fault process, characterized in that, include: The multi-source time-series data module is used to acquire multi-source time-series data of the entire process of power system faults; The data repair module is used to preprocess multi-source time-series data to generate a unified measurement sequence; The frequency dynamic prediction module is used to input the measurement sequence into a pre-constructed fusion factor map and the frequency dynamic estimation model of PINN, and output the dynamic frequency of the entire fault process. The frequency security boundary module is used to extract key frequency security indicators throughout the fault process based on the frequency trajectory and construct the frequency security boundary. The inertia configuration module is used to construct a minimum inertia optimization model based on frequency safety boundary constraints, and output the optimal minimum inertia configuration scheme that satisfies the frequency safety boundary throughout the entire process.

8. A computer-readable storage medium for storing one or more programs, characterized in that: The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1 to 5.

9. A computing device, characterized in that, include: One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods according to claims 1 to 5.

Images