A method and system for power system frequency monitoring based on physical information neural networks

Abstract

This application discloses a power system frequency monitoring method and system based on a physical information neural network, relating to the field of power system stability monitoring. The method includes: constructing a state vector of the power system based on the real-time operating parameters of each unit in the power system to be monitored; predicting the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model, obtaining a frequency prediction trajectory; wherein the power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network; extracting multiple monitoring indicators from the frequency prediction trajectory, and monitoring the frequency of the power system based on the multiple monitoring indicators. By implementing this application, the actual requirements for power system frequency monitoring accuracy in new energy access scenarios can be met.

Description

Technical Field

[0001] This application relates to the field of power system stability monitoring, and in particular to a power system frequency monitoring method and system based on physical information neural networks. Background Technology

[0002] With the deepening construction of new power systems, the installed capacity and power generation share of new energy sources are continuously increasing. Traditional synchronous generator units are being largely replaced by new energy sources with randomness and volatility, leading to a significant decrease in the system's equivalent rotational inertia and a marked weakening of the system's ability to withstand frequency fluctuations, making it susceptible to power system collapse. Therefore, it is necessary to achieve accurate frequency monitoring after system disturbances, including risk perception and trajectory prediction, to ensure the safe and stable operation of the power system under extreme conditions.

[0003] Currently, traditional power system frequency monitoring methods are typically based on preset fixed physical parameters, using numerical simulation or probabilistic statistics for modeling and calculation. While these methods can achieve detailed modeling, the sheer computational complexity in large-scale power systems makes them unsuitable for millisecond-level real-time early warning. Furthermore, they struggle to perceive and align with the real-time physical state of the system, leading to a significant drop in prediction accuracy under parameter mismatch conditions. In recent years, pure data-driven model monitoring methods based on artificial intelligence technologies such as deep learning have emerged. Although they offer the advantage of fast inference speed, they are essentially black-box models lacking physical interpretability. They fail to express the fundamental physical laws of power systems, exhibiting serious errors in predictions for extreme data, such as physical distortion and inconsistencies with common sense, making it difficult to meet the high-precision frequency monitoring requirements of power systems. Therefore, in scenarios involving the integration of renewable energy, how to achieve high-precision frequency monitoring of the power system remains a pressing issue that current technologies must address. Summary of the Invention

[0004] This application provides a power system frequency monitoring method and system based on physical information neural networks to solve the technical problem that the accuracy of existing power system frequency monitoring cannot meet the actual needs in the scenario of new energy access.

[0005] According to a first aspect of the embodiments of this application, a power system frequency monitoring method based on a physical information neural network is provided, comprising: Based on the real-time operating parameters of each unit in the power system to be monitored, the state vector of the power system is constructed; Based on the state vector and combined with a preset power system frequency prediction model, the frequency of the power system within a preset time span is predicted to obtain the frequency prediction trajectory; wherein, the power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. Multiple monitoring indicators are extracted from the frequency prediction trajectory, and the frequency of the power system is monitored based on the multiple monitoring indicators.

[0006] This application first constructs a state vector based on the real-time operating parameters of each unit in the power system, and then combines it with a power system frequency prediction model based on a physical information neural network to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. Multiple monitoring indicators are then extracted from the frequency prediction trajectory and frequency monitoring is implemented. By constructing a power system frequency prediction model through a physical information neural network and using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, the prediction results can be ensured to strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs.

[0007] In some embodiments of this application, the power system frequency prediction model is based on a physical information neural network, and is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, specifically including: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

[0008] This application first models the transient frequency response process based on the power system's system design to obtain a transient frequency response model and determine the operating boundary. Then, it constructs an initial power system frequency prediction model based on a physical information neural network. During construction, the transient frequency response model is converted into physical constraint terms in the total model loss, and the operating boundary is converted into safety constraint terms in the total model loss. The model is then iteratively trained using historical operating datasets to obtain the power system frequency prediction model. By constructing the power system frequency prediction model through a physical information neural network, and using the power system's transient frequency response process as the corresponding physical constraint term and the power system's operating boundary as the corresponding safety constraint term, this application ensures that the training and prediction results strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs.

[0009] In some embodiments of this application, the step of modeling the transient frequency response process of the power system according to the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

[0010] This application first assumes all generating units in the power system are synchronous units to perform equivalent modeling of the power system through the rotor motion of the synchronous units, obtaining a first frequency response model under power deficit conditions. Then, it models a virtual synchronous unit separately to obtain a second frequency response model in which the output power of the virtual synchronous unit dynamically responds to the frequency changes of the power system. Subsequently, it models the transient frequency response process of the power system based on the principle of dynamic power balance to obtain a transient frequency response model. By assuming all generating units are synchronous units for modeling, the overall frequency response of the power system under non-renewable energy access conditions can be simulated. Then, by modeling the virtual synchronous unit separately, the dynamic response of frequency changes under renewable energy access conditions can be simulated. Finally, by combining the two, the overall frequency response of the power system under renewable energy access conditions can be accurately modeled, providing corresponding physical constraint support for the subsequent construction of a power system frequency prediction model.

[0011] In some embodiments of this application, the step of predicting the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model to obtain a frequency prediction trajectory specifically includes: The state vector is copied and extended along the dimension of the number of generating units in the power system, and then concatenated with a preset generating unit identification matrix to obtain a state-enhanced tensor; wherein, the generating unit identification matrix is ​​constructed based on each generating unit in the power system; The state-enhanced tensor is input into the power system frequency prediction model to predict the frequency of the power system within a preset time span, thus obtaining the frequency prediction trajectory.

[0012] This application first replicates and extends the state vector along the dimension of the number of generating units in the power system, and then concatenates the generating unit identification matrix to obtain the state enhancement tensor. This allows for the identification of different generating units, enhancing the expressive power of the state vector. Subsequently, it is input into the power system frequency prediction model to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. This results in a frequency prediction trajectory that is more closely matched to the current task requirements, improving the accuracy of frequency prediction and thus enhancing the accuracy of subsequent frequency monitoring in new energy access scenarios.

[0013] In some embodiments of this application, the plurality of monitoring indicators includes the maximum rate of frequency change, the lowest frequency point, and the frequency recovery stabilization time; the extraction of the plurality of monitoring indicators from the frequency prediction trajectory specifically includes: Based on the frequency prediction trajectory, the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance is calculated to obtain the maximum frequency change rate of the frequency prediction trajectory. A global minimum search is performed on the frequency prediction trajectory to determine the minimum point of the frequency prediction trajectory, thereby obtaining the lowest frequency point of the frequency prediction trajectory. The frequency recovery stabilization time of the frequency prediction trajectory is determined based on the first moment corresponding to the lowest frequency point and the second moment that is closest to the first moment and whose duration is greater than a preset time threshold.

[0014] This application first calculates the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance to obtain the maximum frequency change rate. Then, it performs a global minimum search on the frequency prediction trajectory to determine the minimum point and obtain the lowest frequency point. Based on the frequency prediction trajectory, it determines the frequency recovery stabilization time from the first time corresponding to the lowest frequency point and the second time that is closest to the first time point and whose duration is greater than a preset time threshold. This can obtain multiple monitoring indicators that match the current task requirements and provide a data foundation for subsequent frequency monitoring.

[0015] In some embodiments of this application, monitoring the frequency of the power system based on the plurality of monitoring indicators specifically includes: Based on the multiple monitoring indicators and combined with multiple preset monitoring thresholds, the frequency security status of the power system is determined; Based on the frequency security status, a corresponding frequency warning signal is generated for the power system to complete the monitoring of the power system's frequency.

[0016] This application first determines the frequency security status of the power system based on multiple monitoring indicators and preset multiple monitoring thresholds. It can accurately find and determine the frequency security status of the power system through multiple monitoring indicators and multiple monitoring thresholds, thereby obtaining a more accurate frequency warning signal when generating frequency warning signals for the power system, thus improving the accuracy of frequency monitoring in the scenario of new energy access.

[0017] In some embodiments of this application, determining the frequency security status of the power system based on the plurality of monitoring indicators and in combination with a plurality of preset monitoring thresholds specifically includes: If the maximum frequency change rate is not higher than the preset frequency change threshold, the lowest frequency point is not lower than the preset low-frequency load reduction threshold, and the frequency recovery stabilization time is not higher than the preset recovery time upper limit, then the frequency security status of the power system is determined to be low risk. If the maximum frequency change rate is higher than the frequency change threshold, or the lowest frequency point is lower than the low-frequency load reduction threshold but not lower than the preset absolute frequency lower limit, then the frequency security status of the power system is determined to be medium risk. If the lowest frequency point is lower than the absolute lower frequency limit, or the frequency recovery stabilization time is higher than the recovery time upper limit, then the frequency security status of the power system is determined to be high risk.

[0018] This application determines the frequency safety status of a power system by comparing the maximum frequency change rate with the frequency change threshold, the lowest frequency point with the low-frequency load shedding threshold, the absolute frequency lower limit, and the frequency recovery stabilization time with the upper limit of the recovery time. This method can accurately locate and determine the frequency safety status of the power system, thereby generating more accurate frequency warning signals for the power system in the future.

[0019] According to a second aspect of the embodiments of this application, a power system frequency monitoring system based on physical information neural network is provided, including a state vector construction module, a frequency trajectory prediction module and a system frequency monitoring module; The state vector construction module is used to construct the state vector of the power system based on the real-time operating parameters of each unit in the power system to be monitored. The frequency trajectory prediction module is used to predict the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model, thereby obtaining a frequency prediction trajectory. The power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. The system frequency monitoring module is used to extract multiple monitoring indicators from the frequency prediction trajectory and monitor the frequency of the power system based on the multiple monitoring indicators.

[0020] In some embodiments of this application, the power system frequency prediction model is based on a physical information neural network, and is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, specifically including: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

[0021] In some embodiments of this application, the step of modeling the transient frequency response process of the power system according to the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

[0022] In some embodiments of this application, the frequency trajectory prediction module includes a vector enhancement unit and a frequency prediction unit; The vector enhancement unit is used to copy and extend the state vector along the dimension of the number of generating units in the power system, and concatenate it with a preset generating unit identification matrix to obtain a state enhancement tensor; wherein, the generating unit identification matrix is ​​constructed based on each generating unit of the power system; The frequency prediction unit is used to input the state enhancement tensor into the power system frequency prediction model to predict the frequency of the power system within a preset time span, thereby obtaining the frequency prediction trajectory.

[0023] In some embodiments of this application, the plurality of monitoring indicators include the maximum frequency change rate, the lowest frequency point, and the frequency recovery stabilization time; the system frequency monitoring module includes a first calculation unit, a second calculation unit, and a third calculation unit; The first calculation unit is used to calculate the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance based on the frequency prediction trajectory, so as to obtain the maximum frequency change rate of the frequency prediction trajectory. The second calculation unit is used to perform a global minimum search on the frequency prediction trajectory, determine the minimum point of the frequency prediction trajectory, and obtain the lowest frequency point of the frequency prediction trajectory. The third calculation unit is used to determine the frequency recovery stabilization time of the frequency prediction trajectory based on the first moment corresponding to the lowest frequency point and the second moment that is closest to the first moment and whose duration is greater than a preset time threshold.

[0024] In some embodiments of this application, the system frequency monitoring module includes a status determination unit and a frequency monitoring unit; The state determination unit is used to determine the frequency security status of the power system based on the multiple monitoring indicators and multiple preset monitoring thresholds. The frequency monitoring unit is used to generate a corresponding frequency warning signal for the power system based on the frequency security status, so as to complete the monitoring of the frequency of the power system.

[0025] This application first constructs a state vector based on the real-time operating parameters of each unit in the power system, and then combines it with a power system frequency prediction model based on a physical information neural network to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. Multiple monitoring indicators are then extracted from the frequency prediction trajectory and frequency monitoring is implemented. By constructing a power system frequency prediction model through a physical information neural network and using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, the prediction results can be ensured to strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a power system frequency monitoring method based on a physical information neural network, as shown in some embodiments of this application. Figure 2 This is a block diagram of a power system frequency monitoring system based on a physical information neural network, as shown in some embodiments of this application. Detailed Implementation

[0027] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of this application, and should not be construed as limiting the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments shown in this application without inventive effort are within the protection scope of this application.

[0028] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, unless otherwise explicitly specified, "a plurality of" or "several" means two or more.

[0029] Currently, traditional power system frequency monitoring methods are typically based on preset fixed physical parameters, using numerical simulation or probabilistic statistics for modeling and calculation. While these methods can achieve detailed modeling, the sheer computational complexity in large-scale power systems makes them unsuitable for millisecond-level real-time early warning. Furthermore, they struggle to perceive and align with the real-time physical state of the system, leading to a significant drop in prediction accuracy under parameter mismatch conditions. In recent years, pure data-driven model monitoring methods based on artificial intelligence technologies such as deep learning have emerged. Although they offer the advantage of fast inference speed, they are essentially black-box models lacking physical interpretability. They fail to express the fundamental physical laws of power systems, exhibiting serious errors in predictions for extreme data, such as physical distortion and inconsistencies with common sense, making it difficult to meet the high-precision frequency monitoring requirements of power systems. Therefore, in scenarios involving the integration of renewable energy, how to achieve high-precision frequency monitoring of the power system remains a pressing issue that current technologies must address.

[0030] Based on the above technical background, please refer to Figure 1 This application provides a power system frequency monitoring method based on a physical information neural network, including steps S101 to S103, each step of which is as follows: Step S101: Construct the state vector of the power system based on the real-time operating parameters of each unit in the power system to be monitored.

[0031] Specifically, the real-time operating parameters of each unit in the power system to be monitored include, but are not limited to, active power, frequency deviation, equivalent inertia, and comprehensive regulation coefficient.

[0032] Step S102: Based on the state vector and combined with the preset power system frequency prediction model, predict the frequency of the power system within a preset time span to obtain the frequency prediction trajectory; wherein, the power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network.

[0033] In some embodiments of this application, the power system frequency prediction model is based on a physical information neural network, and is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, specifically including: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

[0034] Specifically, when constructing a transient frequency response model for a power system, it is necessary to model the transient frequency response process of the power system. First, the frequency response mechanisms of power systems mainly composed of traditional synchronous generators and those connected to a single virtual synchronous generator are modeled separately. Then, the models are combined to construct a transient frequency response model of the power system containing the virtual synchronous generator, as follows: (1) Establishing a basic model of frequency response for traditional power systems: Based on the rotor motion equations of synchronous generators, multi-machine systems are equivalent to single-machine systems. Considering power systems dominated by synchronous generators, when encountering power deficits... Under the disturbance, the corresponding system frequency response model (i.e., the first frequency response model) is: ; in, This refers to the power deficit encountered by the system, i.e., the unbalanced power caused by the disturbance; Let be the inertial constant of the synchronous generator. The overall equivalent rotational inertia of the corresponding system; Indicates time The changing real-time frequency value of the system, its derivative with respect to time. RoCoF is the rate of change of frequency. For the power generation side at frequency The corresponding output power; For the load side at frequency The corresponding power consumption; This refers to the adjustment coefficient of the synchronous generator speed governor; This is the frequency regulation effect coefficient of the load; The system's rated frequency, This indicates the deviation of the system's real-time frequency from the system's rated frequency.

[0035] (2) Deriving the dynamic response of the output power of the virtual synchronous generator (VSG) to frequency changes: Based on the virtual inertia and virtual damping parameters of the VSG, and based on the principle that the VSG participates in grid regulation by simulating the rotor motion equation and governor characteristics of a synchronous generator, the dynamic mathematical expression of the VSG output power change when the system frequency fluctuates is derived (i.e., the second frequency response model): ; in, The virtual rotational inertia of a single VSG; The system angular frequency, Indicates the system's angular frequency as a function of time. The rate of change; Reference power; This refers to the actual output power of the VSG; This is the system's rated angular frequency; The virtual damping coefficient for a single VSG; This indicates the deviation of the system's real-time angular frequency from the system's rated angular frequency.

[0036] (3) Consider a power system containing multiple VSGs, in the event of a power disturbance When the total frequency response of the system is the result of the combined action of the traditional synchronous generators and all VSGs, the transient frequency response model of the power system can be derived by aggregating the physical characteristics of the traditional synchronous generators and the virtual control characteristics of the VSGs, based on the principle of dynamic power balance: ; in, This represents the total equivalent inertia of the system; The total physical inertia of the traditional synchronous unit represents the physical inertia provided by the traditional rotating unit of the system; This represents the sum of the virtual inertia of all VSGs connected to the grid. This represents the overall system regulation coefficient; This is the regulation coefficient of the main speed controller for traditional synchronous generator sets, representing the response capability of a system dominated by traditional synchronous generator sets to frequency deviations. The load frequency regulation coefficient of the system represents the active power regulation characteristics generated by load fluctuations with frequency. This is the sum of virtual damping for all grid-connected VSGs, used to simulate the damping torque characteristics of synchronous generators to suppress frequency oscillations.

[0037] Specifically, the operating boundaries of a power system include, but are not limited to: the minimum permissible frequency safety threshold of the system. Its preferred value is 48.5 Hz.

[0038] Specifically, Physics-Informed Neural Networks (PINNs) train traditional neural networks by embedding physical laws into them in the form of partial differential equations, organically combining data-driven approaches and physical constraints. This guides the model to learn and output solutions that conform to physical laws, avoiding the problem of physical distortion caused by not learning physical laws. In this application, the architecture of the Physics-Informed Neural Network (PINN) is prior art known to those skilled in the art, and therefore will not be described in detail here.

[0039] Specifically, when constructing the initial power system frequency prediction model through physical information neural networks, transient frequency response models, and operating boundaries, the design includes three parts: input and output structure, multi-objective loss function, and model training method.

[0040] (1) Input and output structure: In the model input, for models containing For a system with multiple generators, an identity matrix is ​​defined to differentiate the dynamic response characteristics of different generators within the system. Each generator in the unit is uniquely identified, and the identification matrix of the unit is represented as follows: ; Therefore, based on the system's state vector and unit identification matrix The model input can be defined as a tensor. By replicating the state vector along the dimension of the number of units This state enhancement tensor is obtained by concatenating it with the unit identification matrix and performing state augmentation. The state enhancement tensor is represented as: ; in, These are the feature dimensions of the state vector and the unit identification matrix, respectively; through row-by-row copying operators Alignment with column vectors of all 1s Achieve state vector replication along the unit number dimension This is done through the matrix concatenation operator. To concatenate the left and right matrices along the feature dimension; In the model output, since the physical information neural network adopts a fully connected feedforward structure, the model output should represent the power system at the next time step. Frequency prediction value Its formula is defined as ;in This represents a prediction operation using a physical information neural network. The current moment; It is defined as follows.

[0041] (2) Multi-objective loss function: Since the physical information neural network in this application needs to satisfy physical laws and safety constraints while fitting the predicted data, the total loss is defined as a weighted aggregation of the data prediction loss term, the physical constraint term, and the safety constraint term. The functional expression of the total loss is: ; in, Predict the loss term for the data; For physical constraints; For safety constraints; These represent the weighting coefficients of the corresponding loss terms; The data prediction loss item This is used to constrain the deviation between predicted values ​​and actual observed values, specifically: ; in, This represents the total number of training samples / observations; Indicates at time The prediction frequency; Indicates at time The true value of the observed frequency; The physical constraints Based on the transient frequency response model, it is used to force the network output to meet the transient frequency response process of the power system, specifically: ; in, This represents the total number of sampling points used to calculate the physical residual; This represents the total equivalent inertia of the system; Indicates at time The derivative of the predicted frequency with respect to time, i.e. the rate of change of the predicted frequency, reflects the dynamic response speed of the system under disturbance. Indicates at time The system power deficit; This represents the system's equivalent overall regulation coefficient, and is related to the transient frequency response model. They have the same physical meaning. Indicates the system at time 10:00 The deviation of the predicted frequency from the system's rated frequency; The security constraints Built upon operational boundaries, this design enhances the model's ability to detect frequency drop risks and ensures that the network output remains within the system's operational boundaries. Specifically: ; in, This represents the total number of sampling points used to calculate safety constraints; The minimum allowed frequency safety threshold for the system; This term is used to quantify the relative deviation of the predicted frequency from the frequency safety threshold; As the activation function, ensure that it only activates when the prediction frequency is below [a certain value]. The function is activated only when the predicted frequency is within the system's operating boundary range, and the function output is 0 if the predicted frequency is within that range.

[0042] (3) Model training method: The initial power system frequency prediction model is iteratively trained. During the iteration process, the neural network parameters are optimized based on the backpropagation algorithm to reduce the total loss. The frequency prediction model for the power system is obtained by minimizing the loss until it is less than a preset loss threshold.

[0043] This application first models the transient frequency response process based on the power system's system design to obtain a transient frequency response model and determine the operating boundary. Then, it constructs an initial power system frequency prediction model based on a physical information neural network. During construction, the transient frequency response model is converted into physical constraint terms in the total model loss, and the operating boundary is converted into safety constraint terms in the total model loss. The model is then iteratively trained using historical operating datasets to obtain the power system frequency prediction model. By constructing the power system frequency prediction model through a physical information neural network, and using the power system's transient frequency response process as the corresponding physical constraint term and the power system's operating boundary as the corresponding safety constraint term, this application ensures that the training and prediction results strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs.

[0044] In some embodiments of this application, the step of modeling the transient frequency response process of the power system according to the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

[0045] This application first assumes all generating units in the power system are synchronous units to perform equivalent modeling of the power system through the rotor motion of the synchronous units, obtaining a first frequency response model under power deficit conditions. Then, it models a virtual synchronous unit separately to obtain a second frequency response model in which the output power of the virtual synchronous unit dynamically responds to the frequency changes of the power system. Subsequently, it models the transient frequency response process of the power system based on the principle of dynamic power balance to obtain a transient frequency response model. By assuming all generating units are synchronous units for modeling, the overall frequency response of the power system under non-renewable energy access conditions can be simulated. Then, by modeling the virtual synchronous unit separately, the dynamic response of frequency changes under renewable energy access conditions can be simulated. Finally, by combining the two, the overall frequency response of the power system under renewable energy access conditions can be accurately modeled, providing corresponding physical constraint support for the subsequent construction of a power system frequency prediction model.

[0046] In some embodiments of this application, the step of predicting the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model to obtain a frequency prediction trajectory specifically includes: The state vector is copied and extended along the dimension of the number of generating units in the power system, and then concatenated with a preset generating unit identification matrix to obtain a state-enhanced tensor; wherein, the generating unit identification matrix is ​​constructed based on each generating unit in the power system; The state-enhanced tensor is input into the power system frequency prediction model to predict the frequency of the power system within a preset time span, thus obtaining the frequency prediction trajectory.

[0047] Specifically, the steps for augmenting the state vector are the same as those for obtaining the state augmentation tensor during training, and will not be repeated here; the current time step... After the corresponding state-enhanced tensor is input into the power system frequency prediction model, the power system frequency prediction model performs continuous forward inference prediction calculations and directly outputs the future frequency prediction results. Frequency prediction trajectory of the system within seconds ,in .

[0048] This application first replicates and extends the state vector along the dimension of the number of generating units in the power system, and then concatenates the generating unit identification matrix to obtain the state enhancement tensor. This allows for the identification of different generating units, enhancing the expressive power of the state vector. Subsequently, it is input into the power system frequency prediction model to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. This results in a frequency prediction trajectory that is more closely matched to the current task requirements, improving the accuracy of frequency prediction and thus enhancing the accuracy of subsequent frequency monitoring in new energy access scenarios.

[0049] Step S103: Extract multiple monitoring indicators from the frequency prediction trajectory, and monitor the frequency of the power system based on the multiple monitoring indicators.

[0050] In some embodiments of this application, the plurality of monitoring indicators includes the maximum rate of frequency change, the lowest frequency point, and the frequency recovery stabilization time; the extraction of the plurality of monitoring indicators from the frequency prediction trajectory specifically includes: Based on the frequency prediction trajectory, the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance is calculated to obtain the maximum frequency change rate of the frequency prediction trajectory. A global minimum search is performed on the frequency prediction trajectory to determine the minimum point of the frequency prediction trajectory, thereby obtaining the lowest frequency point of the frequency prediction trajectory. The frequency recovery stabilization time of the frequency prediction trajectory is determined based on the first moment corresponding to the lowest frequency point and the second moment that is closest to the first moment and whose duration is greater than a preset time threshold.

[0051] Specifically, frequency prediction trajectory The maximum rate of change of frequency (RoCoF) is used to quantify the severity of frequency degradation in the initial stage of a disturbance. When extracting and determining RoCoF, the automatic differentiation function of the power system frequency prediction model can be directly used to solve for the output function. Regarding time The partial derivative is used to calculate the value at the start of the disturbance. The derivative value at that point is used to obtain the maximum rate of frequency change. .

[0052] Specifically, frequency prediction trajectory Lowest frequency point Used to assess the severity of frequency drops, for When performing extraction and determination, the output length can be... Seconds, in essence, represent the frequency prediction trajectory of a discrete frequency prediction sequence. Perform a global minimum search to determine its minimum point, thereby obtaining the point of lowest frequency. The corresponding time is .

[0053] Specifically, frequency prediction trajectory Frequency recovery time Used to evaluate the speed at which a system recovers from a frequency drop to the steady-state region, for When performing extraction, it can be determined. Starting from a certain time, search backward along the time axis to find the first time that satisfies the condition. And the moment when the maintenance condition is met for a certain period of time Thus, the frequency recovery settling time is obtained as ;in The system's rated frequency The allowable deviation zone nearby.

[0054] This application first calculates the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance to obtain the maximum frequency change rate. Then, it performs a global minimum search on the frequency prediction trajectory to determine the minimum point and obtain the lowest frequency point. Based on the frequency prediction trajectory, it determines the frequency recovery stabilization time from the first time corresponding to the lowest frequency point and the second time that is closest to the first time point and whose duration is greater than a preset time threshold. This can obtain multiple monitoring indicators that match the current task requirements and provide a data foundation for subsequent frequency monitoring.

[0055] In some embodiments of this application, monitoring the frequency of the power system based on the plurality of monitoring indicators specifically includes: Based on the multiple monitoring indicators and combined with multiple preset monitoring thresholds, the frequency security status of the power system is determined; Based on the frequency security status, a corresponding frequency warning signal is generated for the power system to complete the monitoring of the power system's frequency.

[0056] This application first determines the frequency security status of the power system based on multiple monitoring indicators and preset multiple monitoring thresholds. It can accurately find and determine the frequency security status of the power system through multiple monitoring indicators and multiple monitoring thresholds, thereby obtaining a more accurate frequency warning signal when generating frequency warning signals for the power system, thus improving the accuracy of frequency monitoring in the scenario of new energy access.

[0057] In some embodiments of this application, determining the frequency security status of the power system based on the plurality of monitoring indicators and in combination with a plurality of preset monitoring thresholds specifically includes: If the maximum frequency change rate is not higher than the preset frequency change threshold The lowest frequency point is not lower than a preset low-frequency load reduction threshold. And the frequency recovery stabilization time is not higher than the preset recovery time upper limit. If the frequency security status of the power system is determined to be low risk, it means that the disturbance is minor and the frequency dynamics are within a safe and controllable range.

[0058] If the maximum frequency change rate is higher than the frequency change threshold Or, the lowest frequency point is lower than the low-frequency load reduction threshold but not lower than the preset absolute frequency lower limit. If the frequency security status of the power system is determined to be medium risk, it means that the disturbance is moderate and there are obvious frequency security risks in the system, which need to be addressed.

[0059] If the lowest frequency point is lower than the absolute frequency lower limit Or the frequency recovery stabilization time is higher than the upper limit of the recovery time. If the frequency security status of the power system is determined to be high risk, it means that the system frequency has approached or exceeded the safety boundary, and there is a risk of instability, requiring immediate preparation for emergency response.

[0060] This application determines the frequency safety status of a power system by comparing the maximum frequency change rate with the frequency change threshold, the lowest frequency point with the low-frequency load shedding threshold, the absolute frequency lower limit, and the frequency recovery stabilization time with the upper limit of the recovery time. This method can accurately locate and determine the frequency safety status of the power system, thereby generating more accurate frequency warning signals for the power system in the future.

[0061] Compared to existing technologies, this application first constructs a state vector based on the real-time operating parameters of each unit in the power system, and then combines it with a power system frequency prediction model built on a physical information neural network to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. Multiple monitoring indicators are then extracted from the frequency prediction trajectory and frequency monitoring is implemented. By constructing a power system frequency prediction model through a physical information neural network and using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, the prediction results can be ensured to strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs.

[0062] For a method corresponding to the one described above, please refer to [link to relevant documentation]. Figure 2 This application provides a power system frequency monitoring system based on physical information neural network, including a state vector construction module 210, a frequency trajectory prediction module 220 and a system frequency monitoring module 230. The state vector construction module 210 is used to construct the state vector of the power system based on the real-time operating parameters of each unit in the power system to be monitored. The frequency trajectory prediction module 220 is used to predict the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model, thereby obtaining a frequency prediction trajectory; wherein, the power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. The system frequency monitoring module 230 is used to extract multiple monitoring indicators from the frequency prediction trajectory and monitor the frequency of the power system based on the multiple monitoring indicators.

[0063] In some embodiments of this application, the power system frequency prediction model is based on a physical information neural network, and is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, specifically including: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

[0064] In some embodiments of this application, the step of modeling the transient frequency response process of the power system according to the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

[0065] In some embodiments of this application, the frequency trajectory prediction module 220 includes a vector enhancement unit and a frequency prediction unit; The vector enhancement unit is used to copy and extend the state vector along the dimension of the number of generating units in the power system, and concatenate it with a preset generating unit identification matrix to obtain a state enhancement tensor; wherein, the generating unit identification matrix is ​​constructed based on each generating unit of the power system; The frequency prediction unit is used to input the state enhancement tensor into the power system frequency prediction model to predict the frequency of the power system within a preset time span, thereby obtaining the frequency prediction trajectory.

[0066] In some embodiments of this application, the plurality of monitoring indicators include the maximum frequency change rate, the lowest frequency point, and the frequency recovery stabilization time; the system frequency monitoring module 230 includes a first calculation unit, a second calculation unit, and a third calculation unit; The first calculation unit is used to calculate the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance based on the frequency prediction trajectory, so as to obtain the maximum frequency change rate of the frequency prediction trajectory. The second calculation unit is used to perform a global minimum search on the frequency prediction trajectory, determine the minimum point of the frequency prediction trajectory, and obtain the lowest frequency point of the frequency prediction trajectory. The third calculation unit is used to determine the frequency recovery stabilization time of the frequency prediction trajectory based on the first moment corresponding to the lowest frequency point and the second moment that is closest to the first moment and whose duration is greater than a preset time threshold.

[0067] In some embodiments of this application, the system frequency monitoring module 230 includes a status determination unit and a frequency monitoring unit; The state determination unit is used to determine the frequency security status of the power system based on the multiple monitoring indicators and multiple preset monitoring thresholds. The frequency monitoring unit is used to generate a corresponding frequency warning signal for the power system based on the frequency security status, so as to complete the monitoring of the frequency of the power system.

[0068] This application first constructs a state vector based on the real-time operating parameters of each unit in the power system, and then combines it with a power system frequency prediction model based on a physical information neural network to predict the frequency of the power system within a preset time span, obtaining a frequency prediction trajectory. Multiple monitoring indicators are then extracted from the frequency prediction trajectory and frequency monitoring is implemented. By constructing a power system frequency prediction model through a physical information neural network and using the transient frequency response process of the power system as a physical constraint term of the physical information neural network, the prediction results can be ensured to strictly follow the physical laws of the power system, avoiding the physical distortion problem that may occur in existing deep learning models under extreme data. At the same time, the physical information neural network retains the powerful and rapid reasoning ability of the neural network model, thereby improving the accuracy of frequency monitoring in new energy access scenarios to meet practical needs.

[0069] It should be understood that the system provided in this application corresponds to the aforementioned method. The power system frequency monitoring system based on physical information neural network provided in this application can realize the power system frequency monitoring method based on physical information neural network provided in any embodiment of this application.

[0070] Adaptively, embodiments of this application also provide a computer device and a computer-readable storage medium.

[0071] The computer device includes: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor; The processor executes the computer program to implement a power system frequency monitoring method based on a physical information neural network, as described in this application.

[0072] The computer-readable storage medium stores multiple instructions adapted for loading by a processor to execute a power system frequency monitoring method based on a physical information neural network according to this application.

[0073] The above description represents some embodiments of this application, providing a further detailed explanation of the purpose, technical solution, and beneficial effects of this application. It should be understood that the above-described embodiments of this application should not be construed as limiting this application. In particular, any changes, modifications, equivalent substitutions, and variations made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A power system frequency monitoring method based on a physical information neural network, characterized in that, include: Based on the real-time operating parameters of each unit in the power system to be monitored, the state vector of the power system is constructed; Based on the state vector and combined with a preset power system frequency prediction model, the frequency of the power system within a preset time span is predicted to obtain the frequency prediction trajectory; wherein, the power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. Multiple monitoring indicators are extracted from the frequency prediction trajectory, and the frequency of the power system is monitored based on the multiple monitoring indicators.

2. The power system frequency monitoring method based on a physical information neural network according to claim 1, characterized in that, The power system frequency prediction model is based on a physical information neural network, which is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. Specifically, it includes: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

3. The power system frequency monitoring method based on a physical information neural network according to claim 2, characterized in that, The step of modeling the transient frequency response process of the power system based on the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

4. The power system frequency monitoring method based on a physical information neural network according to claim 1, characterized in that, The step of predicting the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model to obtain a frequency prediction trajectory specifically includes: The state vector is copied and extended along the dimension of the number of generating units in the power system, and then concatenated with a preset generating unit identification matrix to obtain a state-enhanced tensor; wherein, the generating unit identification matrix is ​​constructed based on each generating unit in the power system; The state-enhanced tensor is input into the power system frequency prediction model to predict the frequency of the power system within a preset time span, thus obtaining the frequency prediction trajectory.

5. The power system frequency monitoring method based on a physical information neural network according to claim 1, characterized in that, The multiple monitoring indicators include the maximum frequency change rate, the frequency minimum point, and the frequency recovery stabilization time; the extraction of multiple monitoring indicators from the frequency prediction trajectory specifically includes: Based on the frequency prediction trajectory, the partial derivative of the frequency prediction trajectory with respect to time at the start of the disturbance is calculated to obtain the maximum frequency change rate of the frequency prediction trajectory. A global minimum search is performed on the frequency prediction trajectory to determine the minimum point of the frequency prediction trajectory, thereby obtaining the lowest frequency point of the frequency prediction trajectory. The frequency recovery stabilization time of the frequency prediction trajectory is determined based on the first moment corresponding to the lowest frequency point and the second moment that is closest to the first moment and whose duration is greater than a preset time threshold.

6. The power system frequency monitoring method based on a physical information neural network according to claim 5, characterized in that, The monitoring of the power system frequency based on the multiple monitoring indicators specifically includes: Based on the multiple monitoring indicators and combined with multiple preset monitoring thresholds, the frequency security status of the power system is determined; Based on the frequency security status, a corresponding frequency warning signal is generated for the power system to complete the monitoring of the power system's frequency.

7. The power system frequency monitoring method based on a physical information neural network according to claim 6, characterized in that, The step of determining the frequency security status of the power system based on the multiple monitoring indicators and multiple preset monitoring thresholds specifically includes: If the maximum frequency change rate is not higher than the preset frequency change threshold, the lowest frequency point is not lower than the preset low-frequency load reduction threshold, and the frequency recovery stabilization time is not higher than the preset recovery time upper limit, then the frequency security status of the power system is determined to be low risk. If the maximum frequency change rate is higher than the frequency change threshold, or the lowest frequency point is lower than the low-frequency load reduction threshold but not lower than the preset absolute frequency lower limit, then the frequency security status of the power system is determined to be medium risk. If the lowest frequency point is lower than the absolute lower frequency limit, or the frequency recovery stabilization time is higher than the recovery time upper limit, then the frequency security status of the power system is determined to be high risk.

8. A power system frequency monitoring system based on a physical information neural network, characterized in that, It includes a state vector construction module, a frequency trajectory prediction module, and a system frequency monitoring module; The state vector construction module is used to construct the state vector of the power system based on the real-time operating parameters of each unit in the power system to be monitored. The frequency trajectory prediction module is used to predict the frequency of the power system within a preset time span based on the state vector and a preset power system frequency prediction model, thereby obtaining a frequency prediction trajectory. The power system frequency prediction model is constructed and trained based on a physical information neural network, using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. The system frequency monitoring module is used to extract multiple monitoring indicators from the frequency prediction trajectory and monitor the frequency of the power system based on the multiple monitoring indicators.

9. A power system frequency monitoring system based on a physical information neural network according to claim 8, characterized in that, The power system frequency prediction model is based on a physical information neural network, which is constructed and trained by using the transient frequency response process of the power system as a physical constraint term of the physical information neural network. Specifically, it includes: Based on the system design of the power system, the transient frequency response process of the power system is modeled to obtain the transient frequency response model of the power system, and the operating boundary of the power system is determined according to the system design. Based on the physical information neural network, and combining the transient frequency response model and the operating boundary, an initial power system frequency prediction model is constructed; wherein, the total loss of the initial power system frequency prediction model is represented as a weighted aggregation of data prediction loss terms, physical constraint terms, and security constraint terms; the physical constraint terms are obtained through the transient frequency response model, and the security constraint terms are obtained through the operating boundary. Based on the historical operation dataset of the power system, the initial power system frequency prediction model is iteratively trained. During the iteration, the network parameters of the initial power system frequency prediction model are optimized based on backpropagation until the total loss of the initial power system frequency prediction model is less than a preset loss threshold, and the training is completed to obtain the power system frequency prediction model.

10. A power system frequency monitoring system based on a physical information neural network according to claim 9, characterized in that, The step of modeling the transient frequency response process of the power system based on the system design of the power system to obtain the transient frequency response model of the power system specifically includes: All generating units in the power system are assumed to be synchronous generating units. Based on the rotor motion of the synchronous generating units, the power system is modeled equivalently to construct the first frequency response model of the power system under power deficit conditions. The virtual synchronous generator units in each unit of the power system are modeled separately, and a second frequency response model is constructed to dynamically respond to the frequency changes of the power system by the output power of the virtual synchronous generator units. Based on the principle of dynamic power balance, the transient frequency response process of the power system is modeled by combining the first frequency response model and the second frequency response model, thus obtaining the transient frequency response model of the power system.

Images