Fault transient waveform intelligent generation method for protection test of flexible DC power transmission system

Abstract

The invention discloses a fault transient waveform intelligent generation method for flexible DC power transmission system protection test. The method comprises the steps of obtaining system parameters and fault parameters of a flexible DC power transmission system; performing normalization processing on the system parameters and the fault parameters, and constructing a training waveform data set; analyzing and screening a key frequency band of the fault transient waveform by adopting a frequency band energy ratio, and determining a sparse frequency domain threshold value; fitting amplitude phase parameters of the key frequency band based on a discrete cosine basis function and a nonlinear least square method; constructing an amplitude and phase dual-channel deep neural network model, and mapping the system parameters and the fault parameters to a sparse frequency domain parameter space; and synthesizing a fault transient waveform through a cosine basis function according to the sparse frequency domain parameters output by the deep neural network. According to the method, sparse frequency domain parameters are predicted according to system and fault parameters, and then fault transient waveforms are generated.

Description

Technical Field

[0001] This invention belongs to the field of relay protection technology in electrical engineering, and particularly relates to an intelligent method for generating fault transient waveforms for protection testing of flexible DC transmission systems. Background Art

[0002] Flexible DC transmission technology boasts numerous advantages, including independent active and reactive power control, low harmonic levels, ease of modularization, and no commutation failure, making it highly valuable for long-distance, large-scale renewable energy transmission. However, after a line fault in a flexible DC transmission system, the capacitors of the voltage source converter (MMC) discharge rapidly, causing the short-circuit current to rise to tens of times the rated current within milliseconds, severely threatening the system's safe operation. Therefore, the current industry consensus is that the protection of flexible DC transmission lines needs to complete fault identification and isolation within 6ms, significantly increasing the performance requirements for protection action speed compared to AC protection. Unlike AC protection principles and devices based on power frequency steady-state electrical quantities, flexible DC transmission system protection requires rapid fault identification within and outside the fault zone based on the transient electrical quantity characteristics after a DC line fault, ensuring the timeliness of protection action commands. Throughout the R&D process and lifecycle of flexible DC transmission system line protection, various tests are required before commissioning and at each stage of operation, including multiple functional and performance tests under various conditions, to ensure that the device meets engineering requirements and operational conditions. Flexible DC protection systems that rely on transient electrical quantities for fault identification and tripping judgment require a large number of fault transient electrical quantity waveforms for testing. Therefore, obtaining fault transient waveforms is crucial to ensuring that flexible DC transmission line protection can be tested comprehensively and accurately.

[0003] Flexible DC transmission systems have a complex physical nature with high nonlinearity and multi-timescale changes. Their fault transient behavior originates from the nonlinear switching characteristics of power electronic devices, which manifests as multi-timescale dynamic response coupling at the microsecond to minute level. The transient development process of electrical quantities after a fault is complex and has many influencing factors, which cannot be directly described by simple formulas. Apart from a small number of fault transient waveforms obtained through actual flexible DC transmission projects, the acquisition of fault transient waveforms generally has three methods: theoretical analysis, electromagnetic transient software simulation, and dynamic model test. (1) For the method of obtaining fault transient electrical quantity waveforms through theoretical analysis, existing studies can only obtain simplified fault equivalent circuits and solve transient waveforms for specific system scenarios. Simplification leads to the loss of some transient features, and there are certain differences in the circuit simplification methods and equipment modeling in different research results, resulting in high complexity of transient process analysis, strong targeting, and difficulty in promotion; the calculation of transient electrical quantities of flexible DC transmission systems that reflect traveling wave characteristics is difficult and the calculation time cost is high; (2) For the method of obtaining fault transient electrical quantity waveforms by electromagnetic transient software simulation, taking the commonly used PSCAD / EMTDC and RTDS simulation software as examples, PSCAD / EMTDC has the problem of excessively long simulation time for power electronic equipment models, and RTDS has the problem of expensive real-time simulation equipment and long hardware-in-the-loop model building time. At the same time, the simulation platform and protection device parameter settings are difficult, which requires certain skills of operators to use simulation software, and the electrical quantities that are not concerned by the protection criteria will also occupy computing resources when the simulation software performs fault simulation numerical calculations, resulting in a waste of material resources and time; (3) For dynamic model tests, they need to be carried out in a scaled-down power system built in the laboratory, which is expensive, has low flexibility and universality, and is difficult to promote to different flexible DC transmission systems. Furthermore, highly nonlinear dynamic model structures and parameters are limited, resulting in insufficient generalization ability and a very limited capacity to reflect system operating conditions and fault characteristics. Therefore, under the comprehensive constraints of nonlinear switching sequential response, multi-timescale response coupling, and complex cross-dimensional interactions, the full-condition, full-dynamic, and full-parameter collaborative characterization of flexible DC transmission systems is difficult to obtain explicitly using traditional methods. This makes it difficult to acquire large-scale datasets of high-confidence and high-completeness transient waveforms. Researching new methods for acquiring fault transient waveforms for flexible DC transmission line protection has significant engineering practical value. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an intelligent generation method for fault transient waveforms used in the protection testing of flexible DC transmission systems. This method quantifies the frequency domain characteristics of fault transient waveforms using sparse frequency domain parameters and uses a dual-channel DNN amplitude and phase neural network to predict sparse frequency domain parameters based on system and fault parameters, thereby generating fault transient waveforms.

[0005] To achieve the above objectives, this invention provides an intelligent method for generating fault transient waveforms for protection testing of flexible DC transmission systems, comprising:

[0006] Obtain system parameters and fault parameters of the flexible DC transmission system;

[0007] The system parameters and fault parameters are normalized to construct a training waveform dataset;

[0008] Key frequency bands of fault transient waveforms are screened by frequency band energy proportion analysis to determine the sparse frequency domain threshold.

[0009] The amplitude and phase parameters of the key frequency band are fitted using discrete cosine basis functions and nonlinear least squares method.

[0010] A dual-channel deep neural network model for amplitude and phase is constructed to map the system parameters and fault parameters to a sparse frequency domain parameter space;

[0011] Based on the sparse frequency domain parameters output by the deep neural network, the fault transient waveform is synthesized using cosine basis functions.

[0012] Optionally, the system parameters include single-pole transmission power.

[0013] Optionally, the fault parameters include fault type, fault location, and transition resistance.

[0014] Optional, the frequency band energy proportion analysis process includes:

[0015] The spectral energy distribution is obtained by performing a discrete Fourier transform on the fault transient waveform;

[0016] Calculate the energy percentage within a preset frequency threshold;

[0017] The sparse frequency domain threshold is determined when the energy percentage exceeds a set threshold.

[0018] Optionally, the amplitude and phase dual-channel deep neural network includes independently trained amplitude prediction channels and phase prediction channels, and the network structure of each channel includes a fully connected layer of 512-256-128-64-32-1.

[0019] Optionally, the cosine basis function synthesis process satisfies:

[0020]

[0021] Among them, f s A represents the sampling frequency of the measurement equipment. k With φ k These are the amplitude and phase parameters, respectively.

[0022] Optionally, the sampling frequency of the fault transient waveform is 50kHz and the duration is 8ms.

[0023] Optionally, the method further includes inputting the generated waveform into the traveling wave protection criterion module to verify the accuracy of the waveform-triggered protection action.

[0024] Technical effects of this invention: This invention discloses an intelligent generation method for fault transient waveforms used in the protection test of flexible DC transmission systems. A training dataset is obtained from a double-ended flexible DC transmission model in PSCAD, and the generated fault transient waveforms are verified using current protection criteria. This method effectively generates line fault transient waveforms. The accuracy of the generated waveforms' action results under the positive line sending-end current protection criteria compared to the actual waveforms reaches over 99.25% under different system and fault parameters. Furthermore, the root mean square error of the action time is less than 0.27, and the squared correlation coefficient is greater than 0.94. Moreover, it does not cause false tripping of the current protection in the case of a negative grounding fault. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 This is a flowchart illustrating an intelligent generation method for fault transient waveforms in a flexible DC transmission system protection test according to an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the simulation model structure of the dual-ended flexible DC transmission system according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the preprocessing method for fault transient waveform datasets in an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of the dynamic selection of cosine basis frequency based on FBER in an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of the waveform fitting effect based on sparse frequency domain parameters in an embodiment of the present invention;

[0031] Figure 6 This is a schematic diagram of a fully connected DNN structure according to an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of sparse frequency domain threshold selection based on FBER in an embodiment of the present invention, where (a) is the transient current spectrum analysis of the sending end of a positive ground fault; and (b) is the transient current spectrum analysis of the sending end of a negative ground fault.

[0033] Figure 8This is a violin plot showing the error between the fitted waveform and the true waveform based on sparse frequency domain parameter restoration in an embodiment of the present invention.

[0034] Figure 9 This is a violin plot showing the error between the fault transient waveform and the actual waveform in the waveform generation model of this invention embodiment;

[0035] Figure 10 This is a schematic diagram of the nonlinear fitting process of fault transient waveform using the least squares method in an embodiment of the present invention. Detailed Implementation

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0037] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0038] Because the transient waveforms of faults in flexible DC transmission systems contain rapidly changing nonlinear dynamic characteristics, the generation model needs to possess nonlinear mapping capabilities in high-order tensor spaces. The task of generating fault transient waveforms lacks explicit time-domain excitation signals, and the inputs are highly structured (i.e., the inputs only include system parameters, fault parameters, and other conditional parameters). Furthermore, the transient process has a wide bandwidth and complex frequency domain characteristics. Therefore, traditional intelligent algorithms such as neural networks struggle to autonomously establish a high-dimensional implicit mapping relationship between the conditional parameter space and the transient waveform space. Deep learning neural networks possess strong nonlinear mapping capabilities and adaptability, making them suitable for fitting the complex relationship between fault transient waveforms and their influencing factors, providing a novel solution for batch transient waveform data generation. To address the problem of the lack of historical sequences as time-series excitations and the difficulty in fitting the mapping relationship between highly structured inputs and transient features in the generation of fault transient waveforms in flexible DC transmission systems, this invention studies a fault transient waveform generation technology for protection testing of flexible DC transmission systems based on sparse frequency domain parameter calculation and a dual-channel amplitude and phase DNN. First, the three-level mapping relationship of the flexible DC protection test process is analyzed to clarify the input and output spaces of the fault transient waveform generation problem. Second, for the problem of complex transient time-domain characteristics, a sparse frequency domain screening method based on the frequency band energy ratio is proposed, and the sparse frequency domain parameters are calculated using discrete cosine basis and nonlinear least squares method. Third, the amplitude and phase dual-channel DNN model is used to fit the mapping relationship between the system and fault parameter space and the sparse frequency domain parameter space, so as to predict the sparse frequency domain parameters based on the system and fault parameters, and then generate the fault transient waveform.

[0039] As shown in Figure 1, this embodiment provides a method for intelligent generation of fault transient waveforms for protection testing of flexible DC transmission systems, including:

[0040] Step 1: Create a training waveform database:

[0041] A fault was set in a dual-ended flexible DC transmission system model built in PSCAD / EMTDC electromagnetic transient simulation software. System and fault parameters (system parameters, fault parameters, etc.) and waveform recordings taken 8 ms after the fault were used as the data source for model training. The simulation model design of the flexible DC transmission system is as follows: Figure 2 As shown.

[0042] System and fault parameters can be categorized into two types: system parameter information and fault condition information. Based on different indicators of these two types of information, the dimensions of the system and fault parameter space can be flexibly expanded to study the mapping relationship between system and fault parameters and sparse frequency domain parameters under each dimension. The system and fault parameters defined in this invention are shown in Table 1. System operating conditions include control strategies, voltage levels, etc., and fault conditions include fault location, transition resistance, etc., which can be added or removed in the future according to actual needs.

[0043] Table 1

[0044]

[0045] When the fault transient waveform generation model is for a specific flexible DC system, the control strategy, voltage level, line parameters, etc., in the system parameters are already determined and are no longer used as variables to train the model. This invention uses a small model for a specific flexible DC system as an example to study the deep learning waveform generation method, but theoretically, this invention has universality for various power systems.

[0046] The model of this invention sets up different system and fault parameters to create datasets for several aspects, including fault type, fault location, transition resistance, single-pole transmission power, and protection action. The parameter settings are shown in Table 2. Since the distance between the positive and negative poles of the flexible DC transmission line is relatively far, the case of inter-pole faults is not considered.

[0047] Table 2

[0048]

[0049] After obtaining the complete set of simulated fault transient waveform data, the dataset is divided into four groups according to different fault types and single-pole transmission power. Within each group, single-fault simulation data is further subdivided based on different fault locations and transition resistances. This data preprocessing method preserves the feature independence of single-simulation transient waveform characteristics while retaining the feature correlation of waveforms under similar systems and fault parameters. The data preprocessing method is as follows: Figure 3 As shown.

[0050] In this invention, the simulation sampling frequency is set to 50kHz, and 400 data points of the transient waveform 8ms after the fault are acquired as the simulated fault transient waveform data for the current protection test. Four variables, namely fault type, fault location, transition resistance, and single-pole transmission power, are selected as the four dimensions of the system and fault parameter space. After normalizing all input data, a deep learning training dataset is created.

[0051] This training data creation method can organically combine the system and fault parameter space with the fault transient waveform, while preserving the characteristic independence of a single fault transient waveform.

[0052] Step 2: Dynamic cosine basis fitting:

[0053] The Discrete Fourier Transform (DFT) is one of the most fundamental frequency domain analysis theories in the engineering field, and its mathematical formula is shown in Equation (1). In the formula, N is the length of the discrete time domain signal x[n] (n=0,1,...,N-1), and X[k] (k=0,1,...,N-1) represents the complex amplitude (including amplitude and phase information) of the k-th frequency component.

[0054]

[0055] The DFT transform constructs a bijective mapping of a signal from the time domain to the frequency domain using finite-dimensional orthogonal Fourier bases, enabling any discrete signal to be decomposed into a linear superposition of harmonic components with definite phase and amplitude.

[0056] Although Fourier bases can theoretically describe transient waveforms of any frequency band using trigonometric functions, they still have theoretical limitations when characterizing transient waveforms with implicit transient impulse characteristics. For transient signals with wideband characteristics, when the entire frequency band is expanded, its frequency domain decomposition theoretically requires frequency domain component basis functions with the same number of time domain sampling points. The physical characteristics will be implicit in the entire spectrum, and the engineering complexity of analyzing in the sampled frequency domain is the same as that of analyzing in the time domain. Considering that the frequency components contained in the transient electrical quantities of flexible DC faults are only concentrated in some low-frequency bands, and are basically zero in other frequency bands, this invention first analyzes its main frequency domain distribution range to represent the transient characteristics of the waveform through a sparse frequency domain, simplifying the difficulty of waveform generation.

[0057] To quantify the criteria for frequency domain selection, Frequency Band Energy Ratio (FBER) analysis is used for the sparse frequency domain dynamic selection of fault transient electrical quantities. FBER obtains the amplitude and energy of each component by performing a DFT transform on the fault transient waveform, thus achieving a quantitative calculation of the energy ratio within the target frequency band, as shown in equation (2). In the equation, X(f) is the discrete spectrum obtained after performing a DFT on the transient waveform, and f ≥ 0.s f is the sampling frequency. max The preset frequency threshold is f. s Integer multiples of.

[0058]

[0059] The energy percentage threshold η is set according to engineering requirements. When equation (3) is satisfied, it indicates that the energy of the transient waveform is within f. max The sparse frequency domain, which is sufficiently concentrated within the range, can characterize transient features.

[0060] FBER(f max )≥η(3);

[0061] Based on the bijective property of DFT, it can be represented by positive frequency cosine components that include amplitude and phase information, as shown in equation (4):

[0062]

[0063] Where, f s A represents the sampling frequency of the measurement equipment. k With φ k These are the amplitude and phase parameters, respectively. The amplitude and phase of the cosine component at each frequency in equation (4) can directly reflect the harmonic distribution characteristics in the transient waveform.

[0064] The frequency threshold f in the sparse frequency domain is selected based on the energy proportion requirement. max Then, the transient waveform is fitted using a small number of cosine bases in the sparse frequency domain using the nonlinear least squares method to obtain the cosine basis expression of the waveform.

[0065] like Figure 4 As shown, the cosine basis frequency can be dynamically selected using FBER. The frequency range of the cosine basis is determined based on the frequency domain characteristics of different transient waveforms, expanding the range of interest from 0-f. s / 2 reduced to 0-f max .

[0066] By characterizing the transient features of a waveform using a finite number of key frequencies, the interference of meaningless high-frequency noise on the transient features of the waveform is eliminated, and the increase in computational cost due to redundant parameters is avoided, which can significantly improve the efficiency of waveform generation.

[0067] Step 3: Sparse frequency domain parameter calculation method:

[0068] To transform the physical characteristics of the fault transient waveform into mathematical characteristics and provide data for waveform generation, a sparse frequency domain parameter calculation model of the waveform needs to be constructed based on the cosine basis, as shown in Equation (5). In the equation, f(t,μ) is the transient waveform, A0 is the DC component, and A... k cos(2πfk t+φ k f is the AC component at a certain harmonic frequency. k ={f1,f2,...,f n} represents the frequencies in the sparse frequency domain selected by FBER, where μ = {A0, A1, φ1, ..., A n ,φ n} represents the parameters to be optimized.

[0069]

[0070] Nonlinear fitting of fault transient waveforms based on cosine basis expressions is achieved through the least squares method. The nonlinear least squares method finds a set of model parameters (5), and approximates the optimal parameters through iterative optimization, minimizing the sum of squared residuals (SSR) between the model predictions and actual observations. For a given discrete transient waveform with N data points, the mathematical expression (6) of the optimal parameter space is solved through nonlinear optimization search, as shown in... Figure 10 As shown;

[0071]

[0072] After obtaining the expression for the transient waveform through nonlinear fitting of the dynamic cosine basis, the transient characteristics of the waveform are characterized by the amplitude and phase of the cosine basis at each frequency in the sparse frequency domain. The fitting effect is as follows: Figure 5 As shown.

[0073] from Figure 5 As can be seen, due to the sparse representation of frequency domain features, the fitted waveform loses a small amount of UHF features at the beginning and end. However, considering that these features account for a very small proportion, and that the transient quantity protection action principle of the flexible DC system is unrelated to UHF characteristics, the lack of such information will not affect the protection action result.

[0074] Step 4: Sparse frequency domain parameter prediction based on DNN:

[0075] In the mathematical expression of waveforms, sparse frequency domain parameters contain the effective frequency domain information in the transient waveform of a fault in a flexible DC transmission system, and are an important basis for generating transient waveforms. The sparse frequency domain parameter space is shown in equation (7).

[0076] μ={A0,A1,φ1,...,A n ,φ n} (7);

[0077] In the formula, A k With φ k This represents the phase and amplitude of cosine bases at different frequencies.

[0078] According to the physical relationship of the flexible DC system, the fault transient waveform is affected by system parameters and fault conditions, and the system and fault parameter space is shown in Equation (8).

[0079] ξ={ConStr,Vol,Pos,Res,...} (8);

[0080] In the formula, ConStr, Vol, Pos, and Res represent system and fault parameters such as control strategy, voltage level, fault location, transition resistance, and fault type, respectively.

[0081] System and fault parameters influence the frequency domain parameters of electrical quantities during power system faults through formula constraints corresponding to complex physical models. Meanwhile, sparse frequency domain parameters obtained by fitting using the dynamic cosine basis nonlinear least squares method can describe the frequency domain characteristics; that is, there is a highly nonlinear mapping relationship between the system and fault parameter space and the sparse frequency domain parameter space. Mining the mapping relationship between these two spaces, transforming physical dimension features into mathematical dimension features, is a crucial step in generating transient waveforms under specific system and fault parameters.

[0082] The DNN network structure consists of an input layer, hidden layers, and an output layer. The input layer standardizes system and fault parameters, performing feature dimensionality transformation. The hidden layers extract features through multi-level nonlinear mapping, abstracting them layer by layer. The output layer generates sparse frequency domain parameters based on the system and fault parameter features, achieving parameter mapping from protection testing to the sparse frequency domain. The basic structure of a fully connected DNN is as follows: Figure 6 As shown.

[0083] This invention uses system and fault parameters as input to a DNN and sparse frequency domain parameters as output to train the DNN network. Specifically, it addresses the amplitude A in the cosine basis function. k With phase φ k The physical independence of amplitude and phase is used to determine the dual-channel DNN architecture. The dual-channel architecture independently predicts amplitude and phase, avoiding the cross-interference of fitting errors caused by the same model simultaneously undertaking the prediction tasks of amplitude and phase. At the same time, each sub-network can be updated independently to improve the expressive power and flexibility of the model. The trained DNN can reflect the complex mapping relationship between the system and fault parameter space and the sparse frequency domain parameter space, as shown in Equation (9).

[0084] μ = f DNN (ξ)(9);

[0085] When the system and fault parameter space are determined, the sparse frequency domain parameter space under the condition is predicted by the trained DNN, and the transient waveform of the time domain electrical quantity under the fault condition is restored by the cosine basis.

[0086] Step 5: Waveform quality assessment:

[0087] The goal of generating fault transient waveforms is to make them engineering-worthy and usable for fault analysis and the verification of protection equipment. Therefore, we should not only focus on the overall fit between the generated fault transient waveform and the actual fault transient waveform, but also pay close attention to whether the "fault characteristics" required for protection actions have been successfully generated.

[0088] Therefore, this invention also uses the judgment result of the protection criterion as an evaluation index of the quality of the generated fault transient waveform, thereby quantifying the generation model's ability to generate "fault features" and emphasizing the engineering practical value of the waveform in protection testing.

[0089] Taking the traveling wave protection principle proposed by Siemens as an example, the protection criterion is shown in equation (10).

[0090]

[0091] In the formula, U and I are the line voltage and current, respectively, and Δ i (i = 1, 2, 3) are the setting values ​​for the three criteria. By judging whether the actual waveform and the generated waveform under the same system and fault parameters trigger the protection criteria to produce the same judgment result, and whether the timing of triggering the criteria is close, the waveform generation model is evaluated as to whether it has successfully generated the "fault characteristics" of the transient waveform.

[0092] A specific application example of this invention:

[0093] Using the training data from Step 1 as the data source, a numerical example analysis and result verification were performed on the fault transient waveform generation method for flexible DC protection testing proposed in this invention. A dual-ended flexible DC system was built in PSCAD / EMTDC, and the single-pole transmission power parameter of the control mode in the system parameters, along with the fault type, transition resistance, and fault location in the fault information, were used as variables to generate line current and voltage transient waveforms under various system operating conditions and fault conditions, verifying the model's universality. Parameter settings are shown in Table 2, and data preprocessing methods are as follows. Figure 3 As shown.

[0094] The transient line current of the positive pole under a single-pole ground fault with different combinations of transition resistance and fault location is subjected to DFT transformation, and different FBER thresholds are selected for spectral energy screening. A two-dimensional relationship graph between the proportion of waveforms that meet the FBER threshold requirements and the frequency threshold is plotted, such as... Figure 7 As shown.

[0095] Figure 7In the figure, (a) and (b) respectively show the spectral energy concentration of the transient currents of the positive and negative terminals of the sending end under positive grounding fault. An FBER energy threshold is selected, and the percentage of effective waveforms that meet this energy threshold is calculated at different frequency thresholds. This allows for the selection of frequency thresholds that meet the analysis requirements of most transient waveforms. It can be seen that the higher the energy percentage threshold η in equation (3), the wider the frequency domain to be analyzed. When the sending end positive terminal transient current threshold η is selected as 99.75%, the sparse frequency domain range to be analyzed under positive grounding fault is 0-3000Hz, and the sparse frequency domain range to be analyzed under negative grounding fault is 0-1000Hz. By dynamically selecting the frequency domain range of the cosine basis using FBER, while retaining more than 99.75% of the frequency domain energy information, the original 0-25kHz frequency band (sampling frequency of 50kHz) is reduced to within 3000Hz, reducing the frequency domain calculation by more than 88%, effectively achieving frequency domain sparsity.

[0096] Based on the sparse frequency domain threshold, the cosine basis and nonlinear least squares method will serve as the medium for mapping the amplitude and phase characteristics of transient waveforms from physical space to mathematical space, thereby realizing the calculation of sparse frequency domain parameters.

[0097] The violin plot of the error between the fitted waveform reconstructed from the sparse frequency domain parameters obtained by the nonlinear least squares method and cosine basis and the actual fault transient waveform is shown below. Figure 9 As shown.

[0098] Depend on Figure 8 It can be seen that the fitting error values ​​under positive and negative grounding faults basically conform to a normal distribution, with the distribution mean roughly concentrated around -0.0003 and 0.0002 respectively, both close to 0. The error distribution ranges between 0.0004 and -0.0010, while the current value range of the actual waveform data is between 6.0490 and -0.1157. The fitting error is significantly smaller than the actual current value level, and the fitted waveform based on sparse frequency domain parameter reconstruction can effectively represent the characteristics of the actual waveform.

[0099] A six-layer DNN model was selected as the amplitude and phase dual-channel training model. The DNN model parameter settings are shown in Table 3.

[0100] Table 3

[0101]

[0102] During the training of the amplitude-phase dual-channel DNN model, the dataset is divided into a training set and a test set in an 8:2 ratio. The training set is used for model training, while the test set is not involved in the training process and is used to evaluate the performance of the trained model. The system and fault parameters in the test set are used as input to the DNN model. Based on the sparse frequency domain parameters predicted by the model, a fault transient waveform is generated using a cosine basis, and ε is calculated between this waveform and the actual fault transient waveform under the corresponding conditions. ME The resulting violin diagram with error is shown below. Figure 9 As shown.

[0103] Depend on Figure 9 It can be seen that the prediction error values ​​for both positive and negative grounding faults basically conform to a normal distribution, with a mean close to 0. The error distribution ranges from 0.006 to -0.008, significantly smaller than the actual current value. The error distribution between the generated waveform and the actual waveform is similar to... Figure 9 The error distribution of the waveform reconstructed from the sparse frequency domain parameters is similar to that of the real waveform, indicating that the amplitude and phase dual-channel DNN model can predict the sparse frequency domain parameters well based on the system and fault parameters, and the fault transient waveform generated based on the predicted values ​​of the sparse frequency domain parameters can also fit the morphological characteristics of the real waveform well.

[0104] The fault transient current value is related to the current criterion in the traveling wave protection criterion of equation (12). Based on the current criterion, the model-generated fault transient waveform and the actual fault transient waveform under the same system and fault parameters are judged simultaneously. The protection action results and protection action time of the positive line sending-end circuit breaker are compared to realize the quantitative evaluation of whether the generated waveform contains fault transient characteristics that enable the protection to operate correctly, as shown in Table 4.

[0105] Table 4

[0106]

[0107]

[0108] In Table 4, accuracy refers to whether the generated waveform produces the same protection judgment result as the real waveform; root mean square error and square correlation coefficient measure the time difference between the generated waveform and the real waveform triggering protection action. The smaller the root mean square error value and the closer the square correlation coefficient is to 1, the smaller the protection action time error.

[0109] As shown in Table 4, the accuracy of the generated fault transient waveform's operation results under the positive line sending-end current protection criterion with the actual waveform's operation results exceeds 99.25% under different system and fault parameters. Furthermore, the root mean square error of the operating time is less than 0.27, and the squared correlation coefficient is greater than 0.94. Moreover, it does not cause false tripping of the current protection under negative pole grounding fault conditions. The data indicate that the model proposed in this method has high engineering value.

[0110] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for intelligent generation of fault transient waveforms for protection testing of flexible DC transmission systems, characterized in that, include: Obtain system parameters and fault parameters of the flexible DC transmission system; The system parameters and fault parameters are normalized to construct a training waveform dataset; Key frequency bands of fault transient waveforms are screened by frequency band energy proportion analysis to determine the sparse frequency domain threshold. The amplitude and phase parameters of the key frequency band are fitted using discrete cosine basis functions and nonlinear least squares method. A dual-channel deep neural network model for amplitude and phase is constructed to map the system parameters and fault parameters to a sparse frequency domain parameter space; Based on the sparse frequency domain parameters output by the deep neural network, the fault transient waveform is synthesized using cosine basis functions.

2. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The system parameters include single-pole transmission power.

3. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The fault parameters include fault type, fault location, and transition resistance.

4. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The frequency band energy proportion analysis process includes: The spectral energy distribution is obtained by performing a discrete Fourier transform on the fault transient waveform; Calculate the energy percentage within a preset frequency threshold; When the energy percentage exceeds the set energy threshold, the preset frequency threshold is determined to be the sparse frequency domain threshold.

5. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The amplitude and phase dual-channel deep neural network includes independently trained amplitude prediction channels and phase prediction channels. The network structure of each channel contains a fully connected layer of 512-256-128-64-32-1.

6. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The composition of cosine basis functions satisfies: Among them, f s A represents the sampling frequency of the measurement equipment. k With φ k These are the amplitude and phase parameters, respectively.

7. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The sampling frequency of the fault transient waveform is 50kHz, and the duration is 8ms.

8. The intelligent generation method for fault transient waveforms for protection testing of flexible DC transmission systems as described in claim 1, characterized in that, The method also includes inputting the generated waveform into the traveling wave protection criterion module, and verifying the waveform generation effect based on whether the generated waveform can trigger the same protection action as the real waveform.

Images