Power system dynamic element behavior prediction method and device, electronic equipment and medium

Abstract

The invention relates to the technical field of power system stability analysis and control, in particular to a power system dynamic element behavior prediction method and device, electronic equipment and a medium, and the method comprises the steps: obtaining dynamic response time sequence data of a dynamic element under multiple working conditions; generating a training set with a real behavior track, carrying out end-to-end joint training on the behavior prediction model and the virtual state observer, generating a state space matrix and an observer matrix by using a generative neural network during training, outputting a predicted state track by the observer, inputting the predicted state track and dynamic response data into the behavior prediction model to obtain a predicted behavior track, and outputting the predicted behavior track to the virtual state observer. Synchronously optimizing the two models according to the predicted track and the real track; and predicting a dynamic element behavior track by using the trained behavior prediction model. Therefore, the problems that mechanism modeling depends on accurate parameters, a black box model cannot be explained, an identification method is poor in generalization, a neural differential equation does not define an input and output structure, and the state cannot be measured due to the fact that the neural differential equation cannot be processed are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system stability analysis and control technology, and in particular to a power system dynamic element behavior prediction method and device, an electronic device and a medium. BACKGROUND

[0002] The small signal model of the power system is the basis for stability analysis, oscillation mode identification and controller design. Due to the large number of grid-connected power electronic devices dominated by new energy, the modeling method of the dynamic elements of the power system is difficult to establish a high-precision small signal model suitable for a wide range of operating conditions without relying on accurate physical parameters, and when the internal key state variables of the element cannot be directly measured, it leads to difficulties in model training and application. SUMMARY

[0003] The present application provides a power system dynamic element behavior prediction method, device, electronic device and storage medium to solve the problems of mechanism modeling relying on accurate parameters, black box model being uninterpretable, identification method being poor in generalization, neural differential equation not defining input and output structure and being unable to handle unmeasurable states in related technologies.

[0004] The first aspect of the present application provides a power system dynamic element behavior prediction method, comprising the following steps: obtaining dynamic response time series data of dynamic elements in a power system under a plurality of different operating conditions; generating a training data set carrying a real behavior trajectory according to the dynamic response time series data, and performing end-to-end joint training of a behavior prediction model and a virtual state observer using the training data set. In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated using a generative neural network, a predicted state trajectory is determined according to the output of the virtual state observer, the predicted state trajectory and the dynamic response time series data are input into the behavior prediction model, the behavior prediction model outputs a predicted behavior trajectory, and the behavior prediction model and the virtual state observer are simultaneously optimized according to the predicted behavior trajectory determined by the state space matrix and the real behavior trajectory. The trained behavior prediction model is used to predict the behavior trajectory of the dynamic element.

[0005] Optionally, the dynamic response time series data includes at least one of port input variable time series data, port output variable time series data and operating condition time series data of the dynamic element under a plurality of different operating conditions.

[0006] Optionally, the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer are generated using a generative neural network, comprising: inputting the operating condition time series data into the generative neural network, and the generative neural network generating the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer.

[0007] Optionally, determining the predicted state trajectory according to the output of the virtual state observer comprises: identifying the port input variable time series data and the port output variable time series data in the dynamic response time series data; inputting the port input variable time series data and the port output variable time series data into the virtual state observer, and the virtual state observer outputs an initial state; taking the initial state as an initial value, integrating the state equation of the behavior prediction model by using a numerical differential equation solver, and determining the predicted state trajectory according to the integration result.

[0008] Optionally, the state equation of the behavior prediction model is:

[0009] wherein, is a state variable; is an estimated value of the state variable; is a local operating condition vector; is an input variable; is an output variable; is a state matrix; is an input matrix; is an output matrix; is a feedforward matrix.

[0010] Optionally, the state equation of the virtual state observer is:

[0011] wherein, is the predicted state trajectory; is an observer matrix; is a local operating condition vector; is an augmented vector; is an observer matrix; is an augmented vector; is a bias vector.

[0012] Optionally, inputting the predicted state trajectory and the dynamic response time series data into the behavior prediction model comprises: identifying the port input variable time series data in the dynamic response time series data; and inputting the predicted state trajectory and the port input variable time series data into the behavior prediction model.

[0013] The second aspect embodiment of the present application provides a power system dynamic element behavior prediction device, comprising: an acquisition module configured to acquire dynamic response time series data of a dynamic element in a power system under a plurality of different operating conditions; a training module configured to generate a training data set carrying a real behavior trajectory according to the dynamic response time series data, and perform end-to-end joint training of a behavior prediction model and a virtual state observer using the training data set. In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated using a generative neural network, a predicted state trajectory is determined according to an output of the virtual state observer, the predicted state trajectory and the dynamic response time series data are input into the behavior prediction model, the behavior prediction model outputs a predicted behavior trajectory, and the behavior prediction model and the virtual state observer are simultaneously optimized according to the predicted behavior trajectory determined by the state space matrix and the real behavior trajectory. A prediction module is configured to predict a dynamic element behavior trajectory using the trained behavior prediction model.

[0014] The third aspect embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the power system dynamic element behavior prediction method of the above embodiments.

[0015] The fourth aspect embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement the power system dynamic element behavior prediction method of the above embodiments.

[0016] Therefore, the present application has at least the following beneficial effects: The embodiments of the present application can generate a training set containing real behavior trajectories based on the acquired multi-condition dynamic response data, perform end-to-end joint training of a behavior prediction model and a virtual state observer, generate a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer using a generative neural network, predict a state trajectory according to an observer output, input the predicted state trajectory and dynamic response data into the behavior prediction model to obtain a predicted behavior trajectory, and simultaneously optimize the two models according to the predicted and real trajectories. The behavior prediction model is trained to predict a dynamic element behavior trajectory, thereby realizing high-precision generalization of a single model in a wide operating range, effectively solving the problem that key state variables inside an element cannot be directly measured, improving the applicability and robustness of the model in actual application, and the entire modeling process does not need to rely on accurate physical parameters, having the advantages of end-to-end data driving. Thus, the problems of related art, such as mechanism modeling relying on accurate parameters, black-box model being uninterpretable, poor generalization of identification method, neural differential equation not defining input and output structure and being unable to process unmeasurable states, are solved.

[0017] Additional aspects and advantages of the present application will be made apparent by the following description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and be made clear to the reader after a review of the following description and the accompanying drawings. Figure 1 is a flowchart of a power system dynamic element behavior prediction method provided by an embodiment of the present application; Figure 2 is an architectural schematic diagram of a behavior prediction model provided by an embodiment of the present application; Figure 3 is a block schematic diagram of a power system dynamic element behavior prediction device provided by an embodiment of the present application; Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0019] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which examples of the embodiments are shown, wherein the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0020] In the related art, modeling methods are difficult to cope with the development of new power systems dominated by new energy, mainly including: mechanism modeling methods rely on accurate physical parameters, but new energy equipment parameters are time-varying or confidential, making it difficult to obtain, and model deviation seriously affects the analysis accuracy; “black box” data-driven methods lack interpretability, the model structure is disconnected from the physical and cannot provide a state space model that can be used for eigenvalue analysis or controller design; traditional system identification methods rely on specific disturbances, and the identification results have poor generalization ability and are difficult to adapt to multi-working condition operation; existing neural differential equation methods do not clearly define the input and output structure, and do not solve the problem of unobservable key internal state variables, limiting their application in actual power system modeling.

[0021] Therefore, the present application constructs a training set based on multi-working condition data, trains the behavior prediction model and the virtual state observer end-to-end, dynamically outputs the state space matrix and the observer matrix by the generation network, estimates the state and predicts the trajectory using the observer, optimizes the model in combination with the real trajectory, realizes wide-working condition and high-precision generalization, solves the problem of unobservable state, and has the advantage of data-driven without accurate physical parameters.

[0022] The power system dynamic element behavior prediction method, device, electronic equipment and storage medium provided by the embodiments of the present application are described below with reference to the accompanying drawings. In view of the problem that the related technologies mentioned in the background art cannot establish a high-precision small-signal model for dynamic elements in the power system under a wide range of operating conditions without relying on accurate physical parameters, and the state is not measurable, which leads to difficulties in training and application, the present application provides a power system dynamic element behavior prediction method. In the method, based on obtaining multi-condition dynamic response data and generating a training set containing real behavior trajectories, the behavior prediction model and the virtual state observer are trained end-to-end, the state space matrix and the observer matrix are generated by the generation neural network, the predicted state trajectory is obtained according to the observer output, the behavior prediction model is input with the dynamic response data to obtain the predicted behavior trajectory, and the two models are optimized synchronously according to the predicted and real trajectories. The behavior prediction model is used to predict the dynamic element behavior trajectory, so as to realize high-precision generalization of a single model in a wide range of operating intervals, effectively solve the problem that the key state variables in the element cannot be directly measured, improve the applicability and robustness of the model in actual application, and the entire modeling process does not rely on accurate physical parameters, and has the advantages of end-to-end data driving. Thus, the problem that the related technologies cannot establish a high-precision small-signal model for dynamic elements in the power system under a wide range of operating conditions without relying on accurate physical parameters, and the state is not measurable, which leads to difficulties in training and application, is solved.

[0023] Specifically, Figure 1 A flowchart of a power system dynamic element behavior prediction method provided by an embodiment of the present application is shown in FIG. 1.

[0024] As Figure 1 shown, the power system dynamic element behavior prediction method includes the following steps: In step S101, dynamic response time series data of a dynamic element in a power system under a plurality of different operating conditions is obtained.

[0025] It can be understood that the dynamic element refers to a device in the power system that needs to establish a small-signal model, including wind turbines, photovoltaic inverters and other new energy devices and other power electronic devices; the operating condition refers to the quasi-steady state working condition of the element, which is composed of at least one physical quantity such as active power, reactive power or load level.

[0026] In some embodiments, the dynamic response time series data includes at least one of port input variable time series data, port output variable time series data and operating condition time series data of the dynamic element under a plurality of different operating conditions.

[0027] Specifically, dynamic response time series data of a target power system dynamic element under a plurality of different operating conditions is obtained. Each sample data contains: a local operating condition vector of the element port input variable time series data and port output variable time series data . In the case that part or all of the state variables are measurable, state variable time series data is also included .

[0028] In step S102, training data sets carrying real behavior trajectories are generated according to dynamic response time series data, and a pre-constructed behavior prediction model and a virtual state observer are jointly trained end-to-end using the training data sets. In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated using a generative neural network, a predicted state trajectory is determined according to the output of the virtual state observer, the predicted state trajectory and the dynamic response time series data are input into the behavior prediction model, the behavior prediction model outputs a predicted behavior trajectory, and the behavior prediction model and the virtual state observer are simultaneously optimized according to the predicted behavior trajectory determined by the state space matrix and the real behavior trajectory.

[0029] In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated using a generative neural network, a predicted state trajectory is determined according to the output of the virtual state observer, the predicted state trajectory and the dynamic response time series data are input into the behavior prediction model, the behavior prediction model outputs a predicted behavior trajectory, and the behavior prediction model and the virtual state observer are simultaneously optimized according to the predicted behavior trajectory determined by the state space matrix and the real behavior trajectory.

[0030] It can be understood that, in order to accurately model the nonlinear dynamic behavior of the dynamic elements of the power system under different operating conditions, it is necessary to construct a behavior prediction model capable of self-adaptively adjusting parameters according to operating conditions, for describing the input-output dynamic characteristics of the target element; since the internal state variables of the element cannot be directly measured, it is also necessary to construct a virtual state observer relying on external observable signals, to achieve effective estimation of the initial state.

[0031] In some embodiments, the state equation of the behavior prediction model is:

[0032] wherein, is a state variable; is an estimated value of the state variable; is a local operating condition vector; is an input variable; is an output variable; is a state matrix; is an input matrix; is an output matrix; is a feedforward matrix.

[0033] It can be understood that the state matrix , the input matrix , the output matrix , and the feedforward matrix is not fixed but is dynamically generated by a set of parameter generating neural networks according to the local operating condition vector of the element is dynamically generated. That is, wherein represents the set of parameter generating neural networks, for the network parameters to be optimized.

[0034] It should be noted that the behavior prediction model follows the standard state space form of the LTI (Linear Time-Invariant) system, which normalizes the model structure and effectively generates the state space matrix through parameters.

[0035] In some embodiments, the state equation of the virtual state observer is:

[0036] wherein, is the predicted state trajectory; is the observer matrix; is the local operating condition vector; is the augmented vector; is the observer matrix; is the augmented vector; is the bias vector.

[0037] It can be understood that, and are augmented vectors composed of measurable signals, and the augmented vectors are composed of measurable port inputs and outputs and their at least first-order time derivatives, for enhancing the information dimension, satisfying the linear system observability condition, and thereby realizing the reconstruction of unmeasurable states. The observer matrix , and the bias vector are also dynamically generated by another set of parameter generating neural networks according to the operating condition vector .

[0038] It should be noted that the virtual state observer is used to estimate the complete state vector from the measurable port inputs and outputs and their time derivatives.

[0039] In some embodiments, the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer are generated by the generating neural network, including: inputting the operating condition time series data into the generating neural network, and the generating neural network generates the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer.

[0040] Wherein, the state space matrix refers to a coefficient matrix describing a linear relationship between dynamic element inputs and states, outputs; and the observer matrix refers to a transformation matrix in a virtual state observer for linearly reconstructing unmeasurable state variables from an augmented vector composed of measurable port inputs, outputs and derivatives thereof.

[0041] Specifically, in the process of generating the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer by using the generative neural network, for a training sample, the operating condition is input into each parameter generative neural network to obtain a specific state space matrix and an observer matrix under the operating condition.

[0042] In some embodiments, determining the predicted state trajectory according to the output of the virtual state observer comprises: identifying the port input variable time series data and the port output variable time series data in the dynamic response time series data, inputting the port input variable time series data and the port output variable time series data into the virtual state observer, and outputting an initial state by the virtual state observer; taking the initial state as an initial value, integrating the state equation of the behavior prediction model by using a numerical differential equation solver, and determining the predicted state trajectory according to the integration result.

[0043] It can be understood that the embodiments of the present application can estimate the initial state from the measurable port input and output data based on the virtual state observer and solve the state equation by numerical integration, which can accurately construct the predicted state trajectory and effectively solve the modeling problem caused by the unmeasurable internal state of the dynamic element of the power system.

[0044] Specifically, the initial state is estimated from the measurable input / output data by using the virtual state observer, and the estimated is taken as an initial value to integrate the state equation of the behavior prediction model by using an ODESolve (Ordinary Differential Equation Solver), so as to predict the state trajectory .

[0045] In some embodiments, inputting the predicted state trajectory and the dynamic response time series data into the behavior prediction model comprises: identifying the port input variable time series data in the dynamic response time series data; and inputting the predicted state trajectory and the port input variable time series data into the behavior prediction model.

[0046] Specifically, the predicted state trajectory and the given input trajectory are substituted into the output equation of the behavior prediction model to calculate the predicted output trajectory .

[0047] Further, a loss function is constructed, which includes the error between the predicted state trajectory and the real state trajectory, and the error between the predicted output trajectory and the real output trajectory. Through the backpropagation algorithm, the parameters of the generated neural networks are optimized simultaneously according to the gradient of the loss function .

[0048] Specifically, the input of the behavior prediction model includes: the local operating condition vector , which describes the quasi-steady-state quantity of the current operating state of the element, such as the active / reactive power output of the generator, the load level, etc.; the port input variable time series data , which describes the small perturbation quantity of the electrical quantity of the element interacting with the external power grid, such as the real and imaginary parts of the port voltage of the element. The output of the behavior prediction model includes: the port output variable time series data , which describes the small perturbation quantity of the electrical quantity of the element responding to the external power grid, such as the real and imaginary parts of the current injected into the port of the element; the internal state variable time series data , which describes the variable of the internal dynamic characteristics of the element, such as the rotor angle and speed of the synchronous generator.

[0049] Specifically, as shown in Figure 2 , a single power system dynamic element (labeled as element ) is modeled. The construction of the behavior prediction model includes: a parameter generation network, a virtual state observer implementation, and a forward propagation and training process. In the parameter generation network process, the behavior prediction model contains four main parameter generation neural networks, which correspond to the state space matrix respectively. These networks (labeled as and in the figure) take the local operating condition vector as input. In a preferred embodiment, these networks are MLP (Multilayer Perceptron, Multilayer Perceptron). The output of the network is reshaped into the weight (W) and bias (b) of the corresponding matrix, thereby constituting an affine transformation on the input variable.

[0050] For example, for the state equation , its linearized form is:

[0051] By comparing with the standard state space equation, we have:

[0052] where and are the element values of the neural network according to are computed in real time. The output equation is and are also generated in a similar way.

[0053] In the virtual state observer implementation, the virtual state observer is activated when the state variables are not directly measurable. The virtual state observer also contains a set of parameter generating neural networks that take as inputs and generate the observer matrices , and the bias . To estimate the state, the measurable inputs and outputs are needed. According to the linear system observability theory, the state can be reconstructed linearly from the inputs, outputs and their finite order derivatives. Therefore, the augmented vectors and can contain and . In practice, the derivatives can be approximated by numerical differentiation. Then, the estimated value of the state can be computed using the generated observer matrices.

[0054] During the forward propagation and training process, when the training phase is given a data sample: first, the working condition is input into all parameter generating networks to obtain all matrices under this working condition. Then, the initial state is estimated from the measurable data using the observer, and the initial state and input trajectory are sent into a standard ordinary differential equation numerical solver, such as the Runge-Kutta method. The solver integrates the state equation in a forward manner to obtain the state prediction trajectory over the entire time period. Finally, the output prediction trajectory is calculated using the output equation . The loss function is defined as the difference between the prediction trajectory , and the true trajectory , The mean square error between the entire forward calculation process, including the internal operation of the numerical solver of the ordinary differential equation, is differentiable. Therefore, the automatic differentiation and gradient descent algorithm (such as the Adam (Adaptive Moment Estimation) optimizer) can be used to minimize the loss function to train all the parameters of the neural network. After training, for any new operating condition, the trained model can instantly generate the corresponding physically interpretable state space matrix and accurately predict the dynamic behavior of the element, thereby constructing a new paradigm for data-driven, interpretable, wide-range operating condition-adaptive, and state-unobservable power system dynamic element modeling.

[0055] In step S103, the behavior prediction model after training is used to predict the behavior trajectory of the dynamic element.

[0056] It can be understood that the embodiment of the present application can use the behavior prediction model after training to predict the behavior trajectory of the dynamic element, can instantly generate the corresponding physically interpretable state space matrix according to the new operating condition, and realize accurate prediction of the dynamic behavior of the element, thereby supporting small signal stability analysis and control design of the power system. According to the power system dynamic element behavior prediction method proposed in the embodiment of the present application, based on obtaining multi-condition dynamic response data and generating a training set containing real behavior trajectories, the behavior prediction model and the virtual state observer are trained end-to-end, the state space matrix and the observer matrix of the generated neural network are generated, the state trajectory is predicted according to the observer output, the behavior prediction model is input with the dynamic response data to obtain the predicted behavior trajectory, and the two models are optimized according to the predicted and real trajectories, the behavior prediction model after training is used to predict the behavior trajectory of the dynamic element, thereby realizing high-precision generalization of a single model in a wide range of operating intervals, effectively solving the problem that the key state variable inside the element cannot be directly measured, improving the applicability and robustness of the model in actual application, and the entire modeling process does not need to rely on accurate physical parameters, and has the advantages of end-to-end data driving. Therefore, the problems of related art, such as mechanism modeling relying on accurate parameters, black box model being uninterpretable, identification method being poor in generalization, neural differential equation not defining input and output structure and being unable to process state unobservable, are solved.

[0057] Secondly, the power system dynamic element behavior prediction device according to the embodiment of the present application is described with reference to the accompanying drawings.

[0058] Figure 3 is a block schematic diagram of the power system dynamic element behavior prediction device of the embodiment of the present application.

[0059] As Figure 3As shown, the power system dynamic element behavior prediction device 10 comprises an acquisition module 100, a training module 200 and a prediction module 300.

[0060] The acquisition module 100 is configured to acquire dynamic response time series data of dynamic elements in the power system under a plurality of different operating conditions; the training module 200 is configured to generate a training data set carrying a real behavior trajectory according to the dynamic response time series data, and perform end-to-end joint training of a behavior prediction model and a virtual state observer using the training data set. In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated using a generative neural network, a predicted state trajectory is determined according to an output of the virtual state observer, the predicted state trajectory and the dynamic response time series data are input into the behavior prediction model, the behavior prediction model outputs a predicted behavior trajectory, and the behavior prediction model and the virtual state observer are simultaneously optimized according to the predicted behavior trajectory determined by the state space matrix and the real behavior trajectory. The prediction module 300 is configured to predict a dynamic element behavior trajectory using the trained behavior prediction model.

[0061] In some embodiments, the dynamic response time series data comprises at least one of port input variable time series data, port output variable time series data and operating condition time series data of the dynamic element under a plurality of different operating conditions.

[0062] In some embodiments, the training module 200 is configured to input the operating condition time series data into the generative neural network, and the generative neural network generates the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer.

[0063] In some embodiments, the training module 200 is configured to identify the port input variable time series data and the port output variable time series data in the dynamic response time series data, input the port input variable time series data and the port output variable time series data into the virtual state observer, and the virtual state observer outputs an initial state; and integrate a state equation of the behavior prediction model using a numerical differential equation solver with the initial state as an initial value, and determine a predicted state trajectory according to an integration result.

[0064] In some embodiments, the state equation of the behavior prediction model is:

[0065] wherein, is a state variable; is an estimated value of the state variable; is a local operating condition vector; is an input variable; is an output variable; is a state matrix; is an input matrix; is an output matrix; is a feedforward matrix.

[0066] In some embodiments, the state equation of the virtual state observer is:

[0067] wherein, is a predicted state trajectory; is an observer matrix; is a local operating condition vector; is an augmented vector; is an observer matrix; is an augmented vector; is a bias vector.

[0068] In some embodiments, the training module 200 is configured to: identify port input variable time series data in the dynamic response time series data; and input the predicted state trajectory and the port input variable time series data into the behavior prediction model.

[0069] It should be noted that the foregoing description of the embodiment of the power system dynamic element behavior prediction method is also applicable to the embodiment of the power system dynamic element behavior prediction device, which will not be described here again.

[0070] The power system dynamic element behavior prediction device provided by the embodiment of the present application is based on obtaining multi-condition dynamic response data and generating a training set containing a real behavior trajectory, and end-to-end joint training of a behavior prediction model and a virtual state observer, and a generative neural network generates a state space matrix and an observer matrix thereof, and a predicted state trajectory is predicted according to an observer output, and the dynamic response data is input into the behavior prediction model to obtain a predicted behavior trajectory, and the two models are optimized simultaneously according to the predicted and real trajectories, and the behavior prediction model after training is used to predict a dynamic element behavior trajectory, so as to realize high-precision generalization of a single model in a wide range of operating intervals, effectively solve the problem that key state variables in an element cannot be directly measured, and improve the applicability and robustness of the model in actual application. The entire modeling process does not need to rely on accurate physical parameters, and has the advantages of end-to-end data driving. Thus, the problems in the related art, such as dependence of mechanism modeling on accurate parameters, unexplainability of black box models, poor generalization of identification methods, undefined input and output structures of neural differential equations, and inability to process unmeasurable states, are solved.

[0071] Figure 4 The electronic device provided by the embodiment of the present application is shown in the structural schematic diagram of the electronic device. The electronic device can include: The memory 401, the processor 402, and the computer program stored in the memory 401 and executable on the processor 402.

[0072] The processor 402 implements the power system dynamic element behavior prediction method provided in the above embodiments when executing the program.

[0073] Further, the electronic device further comprises: The communication interface 403 is configured to communicate between the memory 401 and the processor 402.

[0074] The memory 401 is configured to store a computer program executable on the processor 402.

[0075] The memory 401 can include a high-speed RAM (Random Access Memory) memory, and can further include a non-volatile memory, for example, at least one disk memory.

[0076] If the memory 401, the processor 402 and the communication interface 403 are independently implemented, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and complete the communication between each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 4 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0077] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can complete the communication between each other through an internal interface.

[0078] The processor 402 can be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0079] The embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the power system dynamic element behavior prediction method of the above embodiments.

[0080] In the description of the application, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the application. In the description of the application, the illustrative description of the above terms is not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or N embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the description and the features of the different embodiments or examples, without contradiction.

[0081] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise explicitly specified.

[0082] Any process or method descriptions in flow charts or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing the specified logical functions or steps, and the various preferred embodiments of the application include additional or fewer steps or processes, as appropriate, and that the steps or processes described with reference to any flowchart or process diagram can be implemented, at least in part, within the programming of one or more computers or computer systems according to the present application.

[0083] It should be understood that parts of the application can be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, the steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. As in another embodiment implemented in hardware, any of the following technologies known in the art or their combinations can be used: discrete logic circuit with logic gate circuit for implementing logic functions on data signals, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array, field programmable gate array, etc.

[0084] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiments can be instructed by the program to complete the relevant hardware, and the above-mentioned program can be stored in a computer readable storage medium. The program includes one or a combination of steps of the method embodiment when executed.

[0085] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that changes, modifications, substitutions and variations can be made by those skilled in the art without departing from the scope of the present application.

Claims

1. A method of power system dynamic element behavior prediction, characterized by, The method comprises the following steps: acquiring dynamic response time series data of a dynamic element in a power system under a plurality of different operating conditions; generating a training data set carrying a real behavior trajectory according to the dynamic response time series data, performing end-to-end joint training of a behavior prediction model and a virtual state observer using the training data set, generating a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer using a generative neural network during the training process, determining a predicted state trajectory according to an output of the virtual state observer, inputting the predicted state trajectory and the dynamic response time series data into the behavior prediction model, the behavior prediction model outputting a predicted behavior trajectory, and synchronously optimizing the behavior prediction model and the virtual state observer according to the predicted behavior trajectory determined according to the state space matrix and the real behavior trajectory; predicting the behavior trajectory of the dynamic element using the trained behavior prediction model.

2. The power system dynamic element behavior prediction method of claim 1, wherein, The dynamic response time series data comprises at least one of port input variable time series data, port output variable time series data and operating condition time series data of the dynamic element under a plurality of different operating conditions.

3. The power system dynamic element behavior prediction method of claim 2, wherein, The method comprises the following steps: inputting the operating condition time series data into the generative neural network, the generative neural network generating the state space matrix of the behavior prediction model and the observer matrix of the virtual state observer.

4. The power system dynamic element behavior prediction method of claim 2, wherein, The method comprises the following steps: identifying the port input variable time series data and the port output variable time series data in the dynamic response time series data, inputting the port input variable time series data and the port output variable time series data into the virtual state observer, the virtual state observer outputting an initial state; using a numerical differential equation solver to integrate a state equation of the behavior prediction model with the initial state as an initial value, and determining the predicted state trajectory according to an integration result.

5. The power system dynamic element behavior prediction method of claim 4, wherein, The state equation of the behavior prediction model is: wherein, is a state variable; is an estimate of the state variable; is a local operating condition vector; is an input variable; is an output variable; is a state matrix; is an input matrix; is an output matrix; is a feedforward matrix.

6. The method of claim 4, wherein, The state equation of the virtual state observer is: wherein, is a predicted state trajectory; is an observer matrix; is a local operating condition vector; is an augmented vector; is an observer matrix; is an augmented vector; is a bias vector.

7. The method of claim 2, wherein the step of determining the behavior of the power system dynamic element comprises the step of: The method comprises the following steps: identifying the port input variable time series data in the dynamic response time series data; inputting the predicted state trajectory and the port input variable time series data into the behavior prediction model.

8. An electric power system dynamic element behavior prediction device characterized by comprising: The method comprises the following steps: an acquisition module configured to acquire dynamic response time series data of a dynamic element in a power system under a plurality of different operating conditions; The training module is configured to generate a training data set carrying a real behavior trajectory according to the dynamic response time series data, and perform end-to-end joint training on a pre-constructed behavior prediction model and a virtual state observer by using the training data set. In the training process, a state space matrix of the behavior prediction model and an observer matrix of the virtual state observer are generated by using a generative neural network. A predicted state trajectory is determined according to an output of the virtual state observer. The predicted state trajectory and the dynamic response time series data are input into the behavior prediction model. The behavior prediction model outputs a predicted behavior trajectory. The predicted behavior trajectory determined according to the state space matrix and the real behavior trajectory are used to synchronously optimize the behavior prediction model and the virtual state observer. The prediction module is configured to predict the dynamic element behavior trajectory by using the trained behavior prediction model.

9. An electronic device, comprising: The computer program or instructions are executed to implement the power system dynamic element behavior prediction method according to any one of claims 1-7. The computer program or instructions are executed to implement the power system dynamic element behavior prediction method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, ​

Images