A method and system for identifying electromagnetic transient model parameters based on autoencoder

Abstract

The present invention discloses a method and system for identifying electromagnetic transient model parameters based on an autoencoder, which relates to the technical field of power system model measurement and identification. The method comprises: obtaining electromagnetic transient model port voltage and active power response curves in an electromagnetic transient model through electromagnetic transient simulation calculation, and integrating them to generate an electromagnetic transient model parameter-response data sample library; inputting a parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a fitter based on a supervised learning algorithm for training, and outputting a trained parameter-response decoder neural network model; inputting the obtained parameter-response decoder neural network model into a pre-built fitter based on an autoencoder algorithm for training, and outputting a trained response-parameter encoder neural network model; and inputting the complete voltage and active power response curves after disturbance into the trained response-parameter encoder neural network model to obtain estimated values ​​of parameters to be identified in the electromagnetic transient model.

Description

Technical Field

[0001] The present invention relates to the technical field of power system model measurement and identification, and in particular to an electromagnetic transient model parameter identification method and system based on an automatic encoder. Background Art

[0002] Existing research has developed a variety of electromagnetic transient model parameter identification methods, including model-driven methods (direct and indirect) and data-driven methods. Direct model-driven methods, or optimization methods, attempt to construct an analytical mapping between model parameters and responses, then solve the problem directly (least squares method) or iteratively (gradient descent method). Direct methods are efficient for simple system models, but their application in electromagnetic transient model parameter identification is limited by the difficulty of obtaining an accurate analytical mapping.

[0003] Indirect model-driven methods are also based on constructing analytical parameter-response models. They aim to update parameters based on the error between the model response and actual measurements, employing either heuristic or filtering algorithms. However, heuristic algorithms are generally inefficient because they require repeated simulations to calculate all transient processes, which is particularly challenging in electromagnetic transient simulations. Filtering methods based on various Kalman filtering and particle filtering techniques are commonly used for dynamic state estimation. The unknown parameters are treated as augmented states, and both state variables and parameters are estimated simultaneously. However, despite their efficiency, filtering methods are very difficult to construct accurate state-space equations for complex electromagnetic transient models, which is crucial for estimating model responses using filtering methods.

[0004] In recent years, artificial intelligence (AI)-based parameter identification methods have emerged, with supervised learning and reinforcement learning (RL) being the primary approaches. SL typically formulates the parameter identification problem as a regression task, trained on a labeled dataset consisting of model responses (data) and corresponding parameters (labels). While these methods promise real-time and efficient parameter identification, they suffer from a fundamental limitation: model nonlinearity can lead to a one-to-many mapping between model responses and parameters, resulting in nonconvergence or failure of parameter identification. Similar to indirect methods, RL typically formulates the parameter correction problem as a Markov decision process. Reinforcement learning techniques are used to train an agent, learning the optimal corrections to the model parameters through repeated interactions between the agent and the environment (i.e., the electromagnetic transient model). After training, the agent is able to adjust the model parameters online. However, due to the typically high computational requirements of electromagnetic transient simulations, the application of reinforcement learning to electromagnetic transient model parameter identification is limited. Summary of the Invention

[0005] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide an electromagnetic transient model parameter identification method and system based on an autoencoder, which can achieve online identification of electromagnetic transient model parameters with both high computational efficiency and identification accuracy.

[0006] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a method for identifying electromagnetic transient model parameters based on an autoencoder, the method comprising the following steps:

[0007] Receiving the parameters of the electromagnetic transient model and the disturbance parameters, respectively obtaining the electromagnetic transient model port voltage and active power response curves corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculations, and integrating the electromagnetic transient model port voltage and active power response curves to generate an electromagnetic transient model parameter-response data sample library;

[0008] Inputting a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-established fitter based on a supervised learning algorithm for training, and outputting a trained parameter-response decoder neural network model;

[0009] Inputting the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, the trained response-parameter encoder neural network model is output;

[0010] The complete voltage and active power response curves after the disturbance are received, and the complete voltage and active power response curves after the disturbance are input into the trained response-parameter encoder neural network model, and the estimated values ​​of the parameters to be identified of the electromagnetic transient model are obtained as output.

[0011] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: obtaining the parameters of the electromagnetic transient model and the disturbance parameters of the electromagnetic transient model to be identified by randomly sampling the parameters using a Monte Carlo method.

[0012] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the electromagnetic transient model parameter-response data sample library includes: N different samples (θ i , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a i The response curves of port voltage V and active power P at time t=1, 2,…, T obtained by electromagnetic transient simulation calculation of the model parameters.

[0013] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established parameter-response decoder neural network model The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model.

[0014] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: inputting the pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on a supervised learning algorithm for training, and outputting the trained parameter-response decoder neural network model as follows:

[0015] The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y corresponding to sample i i The mean square error is minimized, and the training objective is expressed as:

[0016]

[0017] in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library.

[0018] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the pre-established response-parameter encoder neural network model g ψ The input of (·) is the total T-time response curve Y=[V t＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model.

[0019] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: inputting the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on an autoencoder algorithm for training, and outputting the trained response-parameter encoder neural network model as follows:

[0020] The training goal of the autoencoder algorithm is to adjust the response-parameter encoder neural network model g ψ (·) parameter ψ, so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y corresponding to sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as:

[0021]

[0022] Among them, ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library.

[0023] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: inputting the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and outputting the estimated values ​​of the parameters to be identified of the electromagnetic transient model as follows:

[0024] The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

[0025]

[0026] In a second aspect, in order to achieve the above-mentioned purpose, the present invention discloses an electromagnetic transient model parameter identification system based on an autoencoder, comprising:

[0027] A data integration module is used to receive the parameters of the electromagnetic transient model and the disturbance parameters, obtain the electromagnetic transient model port voltage and active power response curve corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculation, and integrate the electromagnetic transient model port voltage and active power response curve to generate an electromagnetic transient model parameter-response data sample library;

[0028] A first training module is used to input a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-built fitter based on a supervised learning algorithm for training, and output a trained parameter-response decoder neural network model;

[0029] The second training module is used to input the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, and output the trained response-parameter encoder neural network model;

[0030] The parameter identification module is used to receive the complete voltage and active power response curves after the disturbance, input the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and output the estimated values ​​of the parameters to be identified of the electromagnetic transient model.

[0031] In conjunction with the second aspect, in certain implementations of the second aspect, the system further comprises: obtaining the parameters of the electromagnetic transient model and the disturbance parameters in the data integration module by randomly sampling the parameters of the electromagnetic transient model to be identified using a Monte Carlo method;

[0032] Or the electromagnetic transient model parameter-response data sample library in the data integration module includes: N different samples (θ i , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a i The response curves of port voltage V and active power P at time t=1, 2, …, T obtained by electromagnetic transient simulation calculation of the model parameters;

[0033] Preferably, the parameter-response decoder neural network model pre-established in the first training module The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model;

[0034] Preferably, in the first training module, the pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library are input into a pre-established fitter based on a supervised learning algorithm for training, and the process of outputting the trained parameter-response decoder neural network model is as follows:

[0035] The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y corresponding to sample i i The mean square error is minimized, and the training objective is expressed as:

[0036]

[0037] in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library;

[0038] Preferably, the response-parameter encoder neural network model g pre-established in the second training module ψ The input of (·) is the total T-time response curve Y=[V t＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model;

[0039] Preferably, in the second training module, the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library are input into a pre-built fitter based on the autoencoder algorithm for training, and the process of outputting the trained response-parameter encoder neural network model is as follows:

[0040] The training goal of the autoencoder algorithm is to adjust the parameter ψ of the response-parameter encoder neural network model gψ(·) so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y in the corresponding sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as:

[0041]

[0042] Where ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library;

[0043] Preferably, the parameter identification module inputs the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and outputs the estimated values ​​of the parameters to be identified of the electromagnetic transient model as follows:

[0044] The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input the trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

[0045]

[0046] Beneficial effects of the present invention:

[0047] Aiming at the problem of online identification of parameters of complex electromagnetic transient models, the present invention adopts an autoencoder structure to prioritize the training of parameter-response encoder neural network model based on traditional supervised learning, and uses it to guide the training of response-parameter decoder neural network model to overcome the "one-to-many" mapping challenge faced by traditional supervised learning methods. While meeting the high computational efficiency requirements of online identification, the accuracy of online identification of electromagnetic transient model parameters is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0049] Figure 1 It is a schematic flow chart of the method of the present invention;

[0050] Figure 2 Schematic diagram of the response-parameter identification model training method based on the autoencoder structure of the present invention;

[0051] Figure 3 Schematic diagram of the training convergence of the parameter-response encoder and the response-parameter decoder of the present invention;

[0052] Figure 4 Schematic diagram comparing the actual measured response curve and the identification parameter simulation response of the present invention;

[0053] Figure 5 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] Example 1:

[0056] The following is an introduction to the relevant terms involved in the embodiments of this application:

[0057] Monte Carlo: The Monte Carlo method is an important numerical analysis method. Its core principle is to approximate the solution to the target problem by performing random sampling in the probability space;

[0058] The basic principle of the Monte Carlo method is to construct samples through random sampling and establish the probability space in which the samples reside, thereby approximating the solution to the target problem. Sampling is a crucial step. Random sampling methods include equal probability sampling, importance sampling, Metropolis sampling, Gibbs sampling, and others. During the problem-solving process, an approximate solution can be calculated through a large number of samplings.

[0059] An autoencoder (AE) is a type of artificial neural network (ANN) used in semi-supervised learning and unsupervised learning. Its function is to perform representation learning on the input information by taking the input information as the learning target.

[0060] An autoencoder consists of two parts: an encoder and a decoder. Based on the learning paradigm, autoencoders can be categorized into contractive autoencoders, regularized autoencoders, and variational autoencoders (VAEs). The first two are discriminative models, while the latter is a generative model. Based on the architecture type, autoencoders can be either feedforward or recursive neural networks.

[0061] like Figure 1 As shown, a method for identifying electromagnetic transient model parameters based on an autoencoder comprises the following steps:

[0062] Receiving the parameters of the electromagnetic transient model and the disturbance parameters, respectively obtaining the electromagnetic transient model port voltage and active power response curves corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculations, and integrating the electromagnetic transient model port voltage and active power response curves to generate an electromagnetic transient model parameter-response data sample library;

[0063] The parameters of the electromagnetic transient model and the disturbance parameters are obtained by randomly sampling the electromagnetic transient model whose parameters are to be identified using the Monte Carlo method;

[0064] The electromagnetic transient model parameter-response data sample library includes: N different samples (θ i , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a iThe response curves of port voltage V and active power P at time t=1, 2, …, T obtained by electromagnetic transient simulation calculation of the model parameters;

[0065] Inputting a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-established fitter based on a supervised learning algorithm for training, and outputting a trained parameter-response decoder neural network model;

[0066] Among them, the pre-established parameter-response decoder neural network model The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model;

[0067] The pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library are input into a pre-established fitter based on a supervised learning algorithm for training. The process of outputting the trained parameter-response decoder neural network model is as follows:

[0068] The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y corresponding to sample i i The mean square error is minimized, and the training objective is expressed as:

[0069]

[0070] in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library;

[0071] Inputting the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, the trained response-parameter encoder neural network model is output;

[0072] Among them, the pre-established response-parameter encoder neural network model g ψ The input of (·) is the total T-time response curve Y=[V t＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model;

[0073] Furthermore, the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library are input into a pre-built fitter based on the autoencoder algorithm for training. The process of outputting the trained response-parameter encoder neural network model is as follows:

[0074] The training goal of the autoencoder algorithm is to adjust the response-parameter encoder neural network model g ψ (·) parameter ψ, so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y in the corresponding sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as:

[0075]

[0076] Among them, ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library.

[0077] The complete voltage and active power response curves after the disturbance are received, and the complete voltage and active power response curves after the disturbance are input into the trained response-parameter encoder neural network model, and the estimated values ​​of the parameters to be identified of the electromagnetic transient model are obtained as output.

[0078] The complete voltage and active power response curves after the disturbance are input into the trained response-parameter encoder neural network model, and the process of outputting the estimated values ​​of the parameters to be identified in the electromagnetic transient model is as follows:

[0079] The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input the trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

[0080]

[0081] Specifically, the present invention will be further described below through examples:

[0082] The specific implementation and effects of the disclosed method are demonstrated based on an electromagnetic transient model of a ±500 kV VSC-HVDC system. This system employs a dq control strategy on both the converter and inverter sides. In this system, the proportional-integral control parameters of both the outer and inner loop controllers are set as target identification objects. The electromagnetic transient modeling and subsequent simulations were performed in the Power System Computer-Aided Design (PSCAD) v4.6 environment. In the electromagnetic transient simulations, the solution and measurement time steps were set to 20 μs / 0.2 ms, and the entire simulation and observation period after the disturbance was set to 5 s / 0.46 s. Using Monte Carlo sampling techniques, a total of 5,000 scenarios were generated, including different unidentified parameters and different three-phase short-circuit fault disturbances on the inverter side. Of these, 4,500 scenarios constituted the training dataset, while the remaining 500 scenarios were used for testing. All parameters related to the model and the disturbance followed a uniform distribution, with the distribution ranges shown in Table 1.

[0083] Table 1 Parameter settings for implementation cases

[0084]

[0085] The training method disclosed in the present invention is used to train the parameter-response encoder and the response-parameter decoder for 400 cycles respectively. Figure 3The mean square error of the fitted parameter-response encoder and response-parameter decoder at each epoch during training is shown. It can be observed that the decoder and encoder converge quickly within the first 50 epochs, stabilizing at approximately 0.005 and eventually converging to approximately 0.0035, highlighting the superior convergence performance of the proposed method. In order to verify the parameter identification performance of the proposed method in its online implementation, a randomly selected test scenario was analyzed in detail. Figure 4 The actual measured response curve is compared with the identified parameter simulation response curve. It can be seen that the parameter simulation curve identified by the trained response-parameter decoder is consistent with the actual response curve.

[0086] Example 2: The second aspect, as Figure 5 As shown, in order to achieve the above purpose, the present invention discloses an electromagnetic transient model parameter identification system based on an automatic encoder, comprising:

[0087] A data integration module is used to receive the parameters of the electromagnetic transient model and the disturbance parameters, obtain the electromagnetic transient model port voltage and active power response curve corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculation, and integrate the electromagnetic transient model port voltage and active power response curve to generate an electromagnetic transient model parameter-response data sample library;

[0088] A first training module is used to input a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-built fitter based on a supervised learning algorithm for training, and output a trained parameter-response decoder neural network model;

[0089] The second training module is used to input the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, and output the trained response-parameter encoder neural network model;

[0090] The parameter identification module is used to receive the complete voltage and active power response curves after the disturbance, input the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and output the estimated values ​​of the parameters to be identified of the electromagnetic transient model.

[0091] In conjunction with the second aspect, in certain implementations of the second aspect, the system further comprises: obtaining the parameters of the electromagnetic transient model and the disturbance parameters in the data integration module by randomly sampling the parameters of the electromagnetic transient model to be identified using a Monte Carlo method;

[0092] Or the electromagnetic transient model parameter-response data sample library in the data integration module includes: N different samples (θi , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a i The response curves of port voltage V and active power P at time t=1, 2, …, T obtained by electromagnetic transient simulation calculation of the model parameters;

[0093] Preferably, the parameter-response decoder neural network model pre-established in the first training module The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model;

[0094] Preferably, in the first training module, the pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library are input into a pre-established fitter based on a supervised learning algorithm for training, and the process of outputting the trained parameter-response decoder neural network model is as follows:

[0095] The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y corresponding to sample i i The mean square error is minimized, and the training objective is expressed as:

[0096]

[0097] in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library;

[0098] Preferably, the response-parameter encoder neural network model g pre-established in the second training module ψ The input of (·) is the total T-time response curve Y=[Vt＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model;

[0099] Preferably, in the second training module, the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library are input into a pre-built fitter based on the autoencoder algorithm for training, and the process of outputting the trained response-parameter encoder neural network model is as follows:

[0100] The training goal of the autoencoder algorithm is to adjust the response-parameter encoder neural network model g ψ (·) parameter ψ, so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y in the corresponding sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as:

[0101]

[0102] Where ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library;

[0103] Preferably, the parameter identification module inputs the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and outputs the estimated values ​​of the parameters to be identified of the electromagnetic transient model as follows:

[0104] The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input the trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

[0105]

[0106] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0107] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which executes the above method when executed by a processor. The storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component.

[0108] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0109] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

Claims

1. A method for identifying electromagnetic transient model parameters based on an autoencoder, characterized in that: The method comprises the following steps: Receiving the parameters of the electromagnetic transient model and the disturbance parameters, respectively obtaining the electromagnetic transient model port voltage and active power response curves corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculations, and integrating the electromagnetic transient model port voltage and active power response curves to generate an electromagnetic transient model parameter-response data sample library; Inputting a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-established fitter based on a supervised learning algorithm for training, and outputting a trained parameter-response decoder neural network model; Inputting the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, the trained response-parameter encoder neural network model is output; The complete voltage and active power response curves after the disturbance are received, and the complete voltage and active power response curves after the disturbance are input into the trained response-parameter encoder neural network model, and the estimated values ​​of the parameters to be identified of the electromagnetic transient model are obtained as output.

2. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The parameters of the electromagnetic transient model and the disturbance parameters are obtained by randomly sampling the electromagnetic transient model whose parameters are to be identified using the Monte Carlo method.

3. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The electromagnetic transient model parameter-response data sample library includes: N different samples (θ i , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a i The response curves of port voltage V and active power P at time t=1, 2,…, T obtained by electromagnetic transient simulation calculation of the model parameters.

4. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The pre-established parameter-response decoder neural network model The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model.

5. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The process of inputting the pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a pre-established fitter based on a supervised learning algorithm for training, and outputting the trained parameter-response decoder neural network model is as follows: The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y in the corresponding sample i i The mean square error is minimized, and the training objective is expressed as: in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library.

6. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The pre-established response-parameter encoder neural network model g ψ The input of (·) is the total T-time response curve Y=[V t＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model.

7. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The process of inputting the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, and outputting the trained response-parameter encoder neural network model is as follows: The training goal of the autoencoder algorithm is to adjust the response-parameter encoder neural network model g ψ (·) parameter ψ, so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y in the corresponding sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as: Among them, ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library.

8. The electromagnetic transient model parameter identification method based on an autoencoder according to claim 1, characterized in that: The process of inputting the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model and outputting the estimated values ​​of the parameters to be identified of the electromagnetic transient model is as follows: The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

9. An electromagnetic transient model parameter identification system based on an autoencoder, characterized in that: include: A data integration module is used to receive the parameters of the electromagnetic transient model and the disturbance parameters, obtain the electromagnetic transient model port voltage and active power response curve corresponding to the electromagnetic transient model parameters and the disturbance parameters through electromagnetic transient simulation calculation, and integrate the electromagnetic transient model port voltage and active power response curve to generate an electromagnetic transient model parameter-response data sample library; A first training module is used to input a pre-established parameter-response decoder neural network model and an electromagnetic transient model parameter-response data sample library into a pre-built fitter based on a supervised learning algorithm for training, and output a trained parameter-response decoder neural network model; The second training module is used to input the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library into a pre-built fitter based on the autoencoder algorithm for training, and output the trained response-parameter encoder neural network model; The parameter identification module is used to receive the complete voltage and active power response curves after the disturbance, input the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and output the estimated values ​​of the parameters to be identified of the electromagnetic transient model.

10. The electromagnetic transient model parameter identification system based on an automatic encoder according to claim 9, characterized in that: The parameters of the electromagnetic transient model and the disturbance parameters in the data integration module are obtained by randomly sampling the parameters of the electromagnetic transient model to be identified using the Monte Carlo method; Or the electromagnetic transient model parameter-response data sample library in the data integration module includes: N different samples (θ i , Y i ),θ i is the parameter of the electromagnetic transient model in sample i of Monte Carlo sampling, Y i =[V i,t＝1,2,…,T ;P i,t＝1,2,…,T ] is a i The response curves of port voltage V and active power P at time t=1, 2, …, T obtained by electromagnetic transient simulation calculation of the model parameters; Preferably, the parameter-response decoder neural network model pre-established in the first training module The input is the model parameter θ and the response Y at the time T0 before the disturbance 0 =[V t＝1,2,…,T0 ;P t＝1,2,…,T0 ], the output is the predicted value Y of all T time response curves pre =[V t＝1,2,…,T ;P t＝1,2,…,T ], are the parameters to be trained of the parameter-response decoder neural network model; Preferably, in the first training module, the pre-established parameter-response decoder neural network model and the electromagnetic transient model parameter-response data sample library are input into a pre-established fitter based on a supervised learning algorithm for training, and the process of outputting the trained parameter-response decoder neural network model is as follows: The training goal of the supervised learning algorithm is to adjust the parameters of the response decoder neural network model parameter So that for each group of samples in the sample library (θ i , Y i ), input electromagnetic transient model parameters θ i The response at the T0 moment before the disturbance After that, the parameter-response decoder neural network model outputs and the response curve Y in the corresponding sample i i The mean square error is minimized, and the training objective is expressed as: in, is the parameter-response decoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library; Preferably, the response-parameter encoder neural network model g pre-established in the second training module ψ The input of (·) is the total T-time response curve Y=[V t＝1,2,…,T ;P t＝1,2,…,T ], the output is the predicted value θ of the electromagnetic transient model parameter to be identified pre , ψ is the parameter to be trained of the response-parameter encoder neural network model; Preferably, in the second training module, the pre-established response-parameter encoder neural network model, the trained parameter-response decoder neural network model, and the electromagnetic transient model parameter-response data sample library are input into a pre-built fitter based on the autoencoder algorithm for training, and the process of outputting the trained response-parameter encoder neural network model is as follows: The training goal of the autoencoder algorithm is to adjust the response-parameter encoder neural network model g ψ (·) parameter ψ, so that for each group of samples (θ i , Y i ), through the autoencoder structure The predicted value of and the response curve Y in the corresponding sample i i The mean square error is the smallest, where the self-encoder structure The predicted value of The calculation process is: to the response-parameter encoder neural network model Enter all T time response curves Y i , and calculate the predicted values ​​of the electromagnetic transient model parameters The response at the T0 moment before the disturbance Input to the trained parameter-response decoder neural network model The predicted values ​​of all T-time response curves are calculated The training objective of the autoencoder algorithm is expressed as: Where ψ* is the response-parameter encoder neural network model parameter after supervised training convergence, and N is the number of samples in the sample library; Preferably, the parameter identification module inputs the complete voltage and active power response curves after the disturbance into the trained response-parameter encoder neural network model, and outputs the estimated values ​​of the parameters to be identified of the electromagnetic transient model as follows: The complete voltage and active power response curve Y after the disturbance obtained by real-time measurement real =[V t＝1,2,…,T ;P t＝1,2,…,T ] Input trained response-parameter encoder neural network model Calculate the estimated value of the electromagnetic transient model parameter θ est , the calculation process is expressed as:

Images