Traveling wave model setting system and method based on waveform transient characteristics

Abstract

The invention discloses a traveling wave model setting system and method based on waveform transient characteristics, and relates to the technical field of power system relay protection and fault localization. Secondary signals output by a traveling wave sensor are decomposed into a plurality of intrinsic mode components; screening effective intrinsic mode components; fitting the waveform of the effective intrinsic mode component, and obtaining a waveform distortion rate and a frequency abrupt change range in real time; adjusting the sampling frequency according to the waveform distortion rate and the frequency abrupt change range; and adopting a deep forest model to output the fault type and position. According to the method, the end effect and mode aliasing of empirical mode decomposition are effectively inhibited through complementary set empirical mode decomposition, the components containing effective traveling wave features can be adaptively reserved in combination with dynamic screening of the multi-sequence permutation entropy on the intrinsic mode components, the signal-to-noise ratio is remarkably improved, the end effect error is reduced, and the method has the advantages of being high in robustness and the like. The defect that effective components cannot be dynamically screened in the prior art is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system relay protection and fault location, and particularly relates to a traveling wave model setting system and method based on waveform transient characteristics. BACKGROUND

[0002] With the development of power systems towards high voltage, large capacity, long distance power transmission, and the high proportion of new energy power generation, the structure of the power grid is becoming increasingly complex, and the probability and harm of faults are also increasing accordingly. As an important part of the power system, the safe and stable operation of the transmission line is directly related to the reliability of the entire power grid. The traveling wave fault location technology has been widely used in transmission line fault detection due to its fast positioning speed and high accuracy.

[0003] Traditional traveling wave positioning methods mainly rely on accurate identification of the wave head. However, in actual power grid environment, the traveling wave signal will be disturbed by various noises in the propagation process, such as line loss, mutual inductor characteristics, external electromagnetic interference, etc., resulting in waveform distortion and wave head characteristics becoming less obvious. In addition, traditional wave head identification methods rely on fixed thresholds or empirical parameters, which are difficult to adapt to complex and variable fault transient processes, and are prone to misjudgment or omission, resulting in large positioning errors, usually about ±300 meters, which is difficult to meet the requirements of modern power grid for fault location accuracy.

[0004] In the prior art, in order to improve the signal-to-noise ratio of the traveling wave signal, empirical mode decomposition (EMD) and other signal decomposition methods are often used, but the EMD method has end effect and mode aliasing problems, which affect the decomposition effect and the accuracy of subsequent feature extraction. In terms of wave head identification, direct detection methods based on sampling points are often used, which are limited by sampling frequency and the discreteness of sampling points, making it difficult to accurately capture the actual arrival time of the wave head. At the same time, a fixed sampling frequency cannot adapt to the rapid changes of the fault transient signal, which may result in the loss of key transient information, further affecting the positioning accuracy. In addition, in fault type identification and positioning calculation, traditional machine learning methods such as support vector machine (SVM) often have difficulty achieving ideal performance in processing small samples, unbalanced data, and complex feature fusion, especially under the changes of power grid characteristics brought by high proportion of new energy, which lacks adaptability.

[0005] Therefore, how to effectively suppress noise interference, accurately extract traveling wave transient characteristics, dynamically adapt to fault transient processes, and achieve high-precision fault location and type identification is a technical problem that needs to be solved in the field. SUMMARY

[0006] The present application aims to overcome the shortcomings of the prior art and provide a traveling wave model setting system and method based on waveform transient characteristics to improve the accuracy of transient waveform feature extraction and fault location accuracy.

[0007] To achieve the above-mentioned objectives, this invention provides a system and method for setting up a traveling wave model based on waveform transient characteristics, comprising:

[0008] First, complementary set empirical mode decomposition (EMD) is used to decompose the secondary signal output by the traveling wave sensor. By adding paired positive and negative Gaussian white noise to the original signal, and then performing multiple EMD decompositions and taking the average value, the noise remaining in the traditional EMD decomposition process is effectively canceled, and the end effect and mode aliasing problems are significantly improved. Thus, the complex traveling wave signal is decomposed into a series of intrinsic mode components with different characteristic scales.

[0009] Furthermore, to filter out components containing effective traveling wave characteristics from the multiple intrinsic mode components obtained from decomposition and eliminate noise-dominated components, this invention introduces multi-sequence permutation entropy. Permutation entropy is a nonlinear dynamic parameter used to measure the complexity of a time series. The lower the entropy value, the stronger the regularity of the signal and the richer the effective information it contains; the higher the entropy value, the more random the signal and the more severe the noise interference. Multi-sequence permutation entropy, by considering the multi-scale characteristics of the signal, can more comprehensively reflect the complexity of the intrinsic mode components. In specific implementation, the permutation entropy value of each intrinsic mode component is calculated, and a reasonable entropy threshold is set, such as 0.3. Intrinsic mode components with entropy values ​​lower than this threshold are determined to be effective components, retained, and further processed, thereby achieving dynamic filtering and enhancement of effective traveling wave characteristics.

[0010] In the precise capture step of wavefront peak, the waveforms of the effective intrinsic mode components obtained in the previous step may still have errors due to the discretization of sampling points. Directly taking the maximum value of the sampling points as the wavefront peak would introduce a large deviation. Therefore, this invention applies a nonlinear least squares optimization algorithm to accurately fit the waveforms of the effective intrinsic mode components. Considering that traveling wavefronts typically have the characteristics of rapid rise and subsequent oscillating decay, this invention innovatively uses a combination function for fitting: for the steep rising edge of the wavefront, a parabolic function is used for fitting, as it can better describe the rapidly changing trend; for the oscillating decay segment after the rising edge, a sine function is used for fitting to characterize its periodic decay characteristics.

[0011] The nonlinear least squares optimization algorithm iteratively optimizes the parameter β to be estimated in the combined function, minimizing the sum of squared residuals between the fitted curve and the actual sampling points of the effective intrinsic mode components, i.e., solving for:

[0012] ;

[0013] yi is the sampled value of the effective intrinsic mode component, f(ti,β) is the combination function of the parabolic function and the sine function, β is the parameter vector to be optimized, and n is the number of sampling points.

[0014] Through this optimized fitting, a continuous and smooth waveform curve can be obtained. Then, by taking the derivative or finding the extreme value of the fitted curve, the time and amplitude of the wavefront peak can be accurately determined, effectively avoiding the errors caused by directly relying on discrete sampling points.

[0015] In the dynamic sampling frequency adjustment step, to ensure that critical signal information is not lost during fault transients, this invention constructs a dynamic sampling frequency adjustment mechanism based on fault characteristic feedback. Traditional fixed sampling frequencies are difficult to adapt to the rapid changes in fault transient signals. When the transient process changes drastically, insufficient sampling rate may lead to waveform distortion; while a continuously high sampling rate can result in excessive data volume and increased processing burden. This invention dynamically adjusts the sampling frequency by analyzing the waveform distortion rate δ and the frequency change range in real time.

[0016] The waveform distortion rate δ is defined as the ratio of the absolute value of the difference between the actual sampled amplitude and the theoretical amplitude to the theoretical amplitude. The theoretical amplitude can be estimated based on historical data from normal operation or a simplified traveling wave propagation model. When δ exceeds a preset threshold, such as 5%, it indicates that the current signal has undergone significant distortion and may contain rich transient information. In this case, the sampling frequency is increased, for example, from 1MHz to 10MHz. When the signal tends to stabilize and the distortion rate decreases below the threshold, the sampling frequency can be appropriately reduced to balance the integrity of data acquisition and system resource consumption. Simultaneously, the frequency fluctuation range of the monitored signal also serves as an auxiliary criterion; when a sudden increase in high-frequency components is detected, the sampling frequency may also be increased.

[0017] In the multimodal data fusion and fault classification process, to comprehensively characterize fault features and improve the accuracy of fault identification and location, this invention extracts rich time-domain and frequency-domain features, totaling 29 dimensions, from the processed traveling wave signal. Time-domain features include, but are not limited to, amplitude abrupt change rate and rise time; frequency-domain features include, but are not limited to, dominant frequency offset and harmonic energy distribution. Specifically, amplitude abrupt change rate reflects the steepness of the wavefront, and rise time is the time from the start of the wavefront to its peak value; the dominant frequency offset in the frequency domain is the difference between the dominant frequency of the signal after the fault and during normal operation, and the harmonic energy distribution is the energy proportion of each harmonic. These multidimensional features characterize the transient properties of the fault from different perspectives.

[0018] To address the issues of high-dimensional features extracted and potential small sample sizes and imbalances in power system fault data, this invention employs a deep forest model for multi-feature fusion and fault classification. Deep forest is a deep learning framework based on decision tree ensembles. It uses a multi-layered structure to progressively abstract features and learn high-level representations by cascading multiple random forests. Compared to traditional deep learning models, deep forest has lower data requirements, is less prone to overfitting, and offers strong interpretability. By weighted fusion of 29-dimensional features, with weights adaptively learned from training data, the deep forest model effectively uncovers nonlinear relationships and combination information between features, ultimately outputting a fault type determination and an accurate estimate of the fault location.

[0019] Corresponding to the method in this invention, this invention also provides a traveling wave model setting system based on waveform transient characteristics, comprising:

[0020] The multimodal signal decomposition and feature enhancement module includes a decomposition unit and a filtering unit; the decomposition unit receives the secondary signal output by the traveling wave sensor and decomposes it; the filtering unit calculates the permutation entropy of each intrinsic modal component and filters out the effective intrinsic modal components.

[0021] The wavefront peak accuracy capture module includes a nonlinear least squares optimization algorithm optimizer and a wavefront calculation unit. The nonlinear least squares optimization algorithm optimizer performs a combination function fitting on the effective intrinsic mode components of the input; the wavefront calculation unit determines the wavefront peak point based on the fitted curve.

[0022] The dynamic sampling frequency control module includes a signal analysis unit and a frequency control unit. The signal analysis unit is used to calculate the distortion rate and monitor frequency abrupt changes, and the frequency control unit adjusts the sampling frequency based on the analysis results.

[0023] The multimodal data fusion and fault classification module includes a feature extraction submodule and a deep forest processing submodule. The feature extraction submodule is used to extract 29-dimensional features, and the deep forest processing submodule is used to load the model, perform fusion calculation, and output the fault type and location.

[0024] Corresponding to the method in this invention, this invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described traveling wave model setting system and method based on waveform transient characteristics.

[0025] Corresponding to the method in this invention, this invention also provides a storage medium storing a computer-readable storage program, which, when executed by a processor, implements the steps of the above-described traveling wave model setting system and method based on waveform transient characteristics.

[0026] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0027] By using complementary set empirical mode decomposition, the endpoint effect and mode aliasing of empirical mode decomposition are effectively suppressed. By combining multi-sequence permutation entropy to dynamically screen intrinsic mode components, components with effective traveling wave characteristics can be adaptively retained, significantly improving the signal-to-noise ratio, reducing endpoint effect errors, and solving the defect of existing technologies that cannot dynamically screen effective components.

[0028] This invention employs a nonlinear least squares optimization algorithm to accurately fit the wavefront waveform, transforming discrete sampling point information into a continuous mathematical model. The peak time of the wavefront is obtained through theoretical calculation, overcoming the limitation of sampling point discreteness and controlling the wavefront timestamp error to ≤0.1μs, thus significantly improving ranging accuracy.

[0029] Secondly, based on real-time monitoring of waveform distortion rate and frequency abrupt changes, this invention achieves adaptive adjustment of the sampling frequency, ensuring the integrity of transient signals and reducing the transient waveform distortion rate from 12% under the traditional fixed sampling rate to 3%. This mechanism can adaptively cope with complex and ever-changing power grid environments, avoiding the loss of key information caused by fixed sampling rates.

[0030] Meanwhile, this invention utilizes a deep forest model to effectively fuse 29-dimensional high-dimensional features, fully mining multifaceted information about fault transients, achieving a fault identification accuracy of 98.5%, which is superior to traditional SVM methods. This model can effectively solve the problem of small sample imbalance and is compatible with complex scenarios such as high-proportion integration of new energy sources. Attached Figure Description

[0031] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0032] Figure 1 A flowchart illustrating the method for setting up a traveling wave model based on waveform transient characteristics. Detailed Implementation

[0033] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0034] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0035] Those skilled in the art should understand that, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting this invention.

[0036] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0037] Example 1

[0038] Please refer to Figure 1 , Figure 1 This invention provides a method for setting up a traveling wave model based on waveform transient characteristics, comprising the following steps:

[0039] Step 1: The secondary signal output by the traveling wave sensor is processed by complementary set empirical mode decomposition, which decomposes it into multiple intrinsic mode components. The effective intrinsic mode components are dynamically screened by multi-sequence permutation entropy.

[0040] The original signal x(t) is given N sets of positive and negative Gaussian white noise ni(t) to generate:

[0041] xi + (t)==x(t)+0.1X ni(t);

[0042] xi - (t)=x(t)-0.1X ni(t);

[0043] Where ni(t) represents Gaussian white noise, and i is the group index to which the noise is added.

[0044] Empirical mode decomposition (EMD) is performed on each signal group to obtain the intrinsic mode components c. + i,j(t) and c - i,j (t);

[0045] The components are averaged:

[0046] Eliminate residual noise.

[0047] Where cj(t) is the j-th intrinsic mode component (after averaging), N is the number of Gaussian white noise groups added (the number of positive and negative noise pairs), and c + Let i,j(t) be the j-th component decomposed after adding positive noise to the i-th group, and c - i,j(t) represents the j-th component decomposed after adding negative noise to the i-th group, and t is the time variable.

[0048] Calculate the multi-sequence permutation entropy (MPE) for the decomposed intrinsic mode components:

[0049]

[0050] Where P(k) is the probability distribution of the symbol sequence after phase space reconstruction.

[0051] Set the threshold MPEth=0.3 to filter out intrinsic modal components that are less than MPEth as valid intrinsic modal components;

[0052] Step 2: Apply a nonlinear least squares optimization algorithm to fit the waveform of the effective intrinsic mode components obtained in Step 1. Specifically, a parabolic function is used to fit the rising edge of the wavefront, and a sine function is used to fit the oscillation decay segment. The peak point of the wavefront is determined through theoretical calculation.

[0053] Functional model:

[0054] ;

[0055] Where tp is the peak time of the wavefront, and β=[β1,…,β7] are the parameters to be optimized.

[0056] Parabolic segment: Characterizes the secondary response characteristics of the traveling wave front affected by the line distribution parameters.

[0057] Sinusoidal attenuation segment: Characterizes the oscillation attenuation caused by line inductance and capacitance to ground.

[0058] Optimization goal:

[0059] ;

[0060] Update parameters using the Jacobian matrix J:

[0061] ;

[0062] When the residual The wavefront time tp is obtained by convergence.

[0063] Where f(t,β) is the function model, t pLet β be the wavefront time, β = [β1, ..., β7], where β contains parabolic coefficients (β1, β2, β3) and sinusoidal attenuation coefficients (β4, β5, β6, β7), yi is the sampled value of the effective intrinsic mode components, J is the Jacobian matrix, r is the residual vector, λ is the damping factor, I is the identity matrix, and k is the number of iterations.

[0064] When the residual <10 -6 The wavefront converges at time tp, and the output wavefront time is tp.

[0065] Step 3: Construct a fault characteristic feedback mechanism to analyze waveform distortion rate and frequency change range in real time; automatically adjust the sampling frequency based on the analysis results to ensure the integrity of transient signals;

[0066] Calculation of distortion rate δ:

[0067]

[0068] In this context, Atheory represents the fundamental amplitude, while Aactual represents the reconstructed amplitude after complementary set empirical mode decomposition.

[0069] When δ>5%, the trigger sampling rate is doubled from 1MHz to 10MHz to ensure that the number of transient signal sampling points is ≥20 cycles.

[0070] Step 4: Extract time-domain and frequency-domain features from the signal processed in Step 1. The time-domain features include amplitude change rate and rise time, and the frequency-domain features include dominant frequency offset and harmonic energy, for a total of 29 features. The extracted 29 features are weighted and fused using a deep forest model to output the fault type and location.

[0071] Time domain: amplitude mutation rate 13 dimensions including rise time (tr) and peak factor (Cf);

[0072] Frequency domain: 16 dimensions including dominant frequency offset Δf, harmonic energy ratio Eh / E0, wavelet entropy Hw, etc.

[0073] Deep forest structure:

[0074] It consists of a 3-layer cascaded random forest, with each layer containing 10 base classifiers, and outputs the fault location through dynamic weighting:

[0075] ;

[0076] Where fault location is the estimated fault location, wm is the weight of the m-th base classifier, fm is the m-th random forest classifier function, f is a 29-dimensional feature vector, and M is the total number of base classes.

[0077] Example 2

[0078] Based on Example 1, the nonlinear least squares optimization algorithm iteratively optimizes the parameter β to be estimated in the combination function, minimizing the sum of squared residuals between the fitted curve and the actual sampling points of the effective intrinsic mode components, i.e., solving for:

[0079] ;

[0080] yi is the sampled value of the effective intrinsic mode component, f(ti,β) is the combination function of the parabolic function and the sine function, β is the parameter vector to be optimized, and n is the number of sampling points.

[0081] Differentiating with respect to β, we get:

[0082]

[0083] By balancing the Levenberg-Marquardt damping factor μ using the Gauss-Newton method and gradient descent, convergence is achieved in 5 iterations, with a wavefront timestamp error ≤ 0.1 μs, corresponding to the transmission line positioning error:

[0084]

[0085] V = 3 × 10 8 m / s, Δt=0.1×10 -6 s;

[0086] The corresponding transmission line positioning error Δl = 15m;

[0087] Example 3

[0088] Based on the above embodiments, the traveling wave model setting system based on waveform transient characteristics includes: a multimodal signal decomposition and feature enhancement module, a wavefront peak accurate capture module, a dynamic sampling frequency control module, and a multimodal data fusion and fault classification module.

[0089] The multimodal signal decomposition and feature enhancement module includes a decomposition unit and a filtering unit; the decomposition unit receives the secondary signal output by the traveling wave sensor and decomposes it; the filtering unit calculates the permutation entropy of each intrinsic modal component and filters out the effective intrinsic modal components.

[0090] The wavefront peak accuracy capture module includes a nonlinear least squares optimization algorithm optimizer and a wavefront calculation unit. The nonlinear least squares optimization algorithm optimizer performs a combination function fitting on the effective intrinsic mode components of the input; the wavefront calculation unit determines the wavefront peak point based on the fitted curve.

[0091] The dynamic sampling frequency control module includes a signal analysis unit and a frequency control unit. The signal analysis unit is used to calculate the distortion rate and monitor frequency abrupt changes, and the frequency control unit adjusts the sampling frequency based on the analysis results.

[0092] The multimodal data fusion and fault classification module includes a feature extraction submodule and a deep forest processing submodule. The feature extraction submodule is used to extract 29-dimensional features, and the deep forest processing submodule is used to load the model, perform fusion calculation, and output the fault type and location.

[0093] Example 4

[0094] Based on the above embodiments, the present invention is used for fault location of 220kV transmission lines.

[0095] The line length is 150km, the fault type is phase A ground fault (45km from the monitoring point), the initial sampling rate is 1MHz, and the traveling wave velocity is 2.98×10⁻⁶. 8 m / s.

[0096] Initial signal: Secondary signal output by traveling wave sensor (A-phase ground fault, 45km from monitoring point).

[0097] Add noise: Inject 20 sets of Gaussian white noise with an amplitude of ±0.1 to generate 40 sets of perturbation signals:

[0098] xi + (t)==x(t)+0.1X ni(t);

[0099] xi - (t)=x(t)-0.1X ni(t);

[0100] Empirical mode decomposition was performed on 40 sets of signals, resulting in a total of 8 intrinsic mode components;

[0101] Averaging eliminates residual noise:

[0102]

[0103] Multiple sequence permutation entropy (MPE) screening:

[0104] Calculate the permutation entropy of each intrinsic mode component, and retain the components with permutation entropy MPE less than 0.3:

[0105] The permutation entropy (MPE) of intrinsic modal components 3-5 are 0.18, 0.22, and 0.25, respectively.

[0106] Intrinsic modal components 3-5 are effective intrinsic modal components.

[0107] Fitting function:

[0108] ;

[0109] Objective: Minimize the sum of squared residuals.

[0110] ;

[0111] Initial parameters: (Default initial value);

[0112] Iterative process:

[0113] Table 1

[0114]

[0115] Output parameters: β=[0.32,15.7,−2.1,4.8,125.6,0.63,12.4];

[0116] Wavehead timing calculation:

[0117] Using the formula for the extreme points of a parabola ;

[0118] t p =-15.7 / 2×0.32=12.658 μs; the theoretical value is 12.658 μs, the actual detection value is 12.738 μs, therefore the error is 0.08 μs.

[0119] Dynamic sampling frequency control:

[0120] 1. Distortion rate monitoring: Theoretical fundamental amplitude A = 2202 / 3 ≈ 179.6 kV;

[0121] The actual reconstructed signal amplitude A is 168.2 kV.

[0122] Distortion rate calculation: δ = |179.6−168.2|÷179.6×100% = 7.5% > 5%;

[0123] 2. Sampling rate adjustment:

[0124] Triggering condition: δ>5%, sampling rate increased from 1 MHz to 10 MHz.

[0125] Continuous monitoring: If the distortion rate drops to δ=1.8%<5% after 3 ms, then resume 1 MHz sampling.

[0126] Fault location and classification:

[0127] 1. Feature Extraction (29 Dimensions)

[0128] Temporal features (13 dimensions):

[0129] Amplitude mutation rate: ;

[0130] Rise time tr = 0.12 μs;

[0131] Peak factor ;

[0132] Frequency domain features (16 dimensions):

[0133] frequency offset ;

[0134] The harmonic energy ratio Eh / E0 = 0.18;

[0135] Wavelet entropy ;

[0136] 2. Deep Forest Model Processing

[0137] Model structure: 3-layer cascaded random forest, with 10 base classifiers in each layer (30 in total).

[0138] Dynamic weighted fusion:

[0139] ;

[0140] Weight allocation: The weights of the previous classifier are wm ∝ 1 / error rate.

[0141] Output result:

[0142] Fault type: Single-phase grounding (phase A) (confidence level 99.2%).

[0143] Fault distance: 45.05 km (theoretical value 45 km, error 0.05 km = 50 m).

[0144] Example 5

[0145] Based on the above embodiments, the multi-sequence permutation entropy threshold is used to filter effective intrinsic mode components, and the waveform distortion rate threshold is used to trigger sampling frequency adjustment. Their values ​​directly affect the signal denoising effect, transient feature extraction accuracy, and system resource consumption. This experiment uses the controlled variable method to analyze the impact of these two parameters at different values ​​on fault location error, transient waveform distortion rate, and fault identification accuracy, verifying the rationality of the existing thresholds (MPE=0.3, distortion rate=5%), and determining the robustness range of the parameters.

[0146] 1. Experimental Environment and Dataset

[0147] Simulation platform: PSCAD / EMTDC was used to build a 220kV transmission line model (150km in length, adjustable sampling rate from 1MHz to 10MHz), simulating 10 typical faults (single-phase grounding, two-phase short circuit, three-phase short circuit, etc.), covering different fault locations (5km to 145km), transition resistance (0.1Ω to 100Ω), and noise levels (signal-to-noise ratio 20dB to 40dB), generating a total of 1000 fault samples.

[0148] Evaluation indicators:

[0149] Location error: The absolute deviation between the actual fault location and the model output location (unit: m);

[0150] Transient waveform distortion rate: The root mean square error between the reconstructed waveform and the ideal waveform under dynamic sampling (unit: %).

[0151] Fault identification accuracy: The percentage of fault types correctly classified by the deep forest model (unit: %).

[0152] 2. MPE Threshold Sensitivity Analysis

[0153] Variable settings: MPE threshold range is 0.2~0.4 (step size 0.05), fixed distortion rate threshold = 5%, other parameters remain unchanged.

[0154] Experimental steps:

[0155] Complementary ensemble empirical mode decomposition is performed on each group of fault samples to calculate the MPE value of each intrinsic mode component;

[0156] Effective intrinsic mode components are selected according to different MPE thresholds, and waveform fitting and feature extraction are performed.

[0157] The mean and standard deviation of the statistical positioning error and recognition accuracy.

[0158] 3. Sensitivity analysis of distortion rate threshold

[0159] Variable settings: The distortion rate threshold range is 3%~7% (step size 1%), the MPE threshold is fixed at 0.3, and other parameters remain unchanged.

[0160] Experimental steps:

[0161] Real-time calculation of waveform distortion rate δ during fault transient process;

[0162] Trigger sampling frequency switching based on different thresholds (The sampling rate was increased from 1MHz to 10MHz when the threshold was reached).

[0163] Statistical transient waveform distortion rate, positioning error, and data storage volume (unit: GB / hour).

[0164] The impact of the MPE threshold on performance is shown in Table 2.

[0165] Table 2

[0166]

[0167] As shown in Table 1, when MPE=0.3, the positioning error is the smallest (48m) and the recognition accuracy is the highest (98.5%). At this time, the number of effective intrinsic mode components is 4.2, which balances feature integrity and noise reduction effect.

[0168] If the threshold is too low (<0.25), excessive removal of high-frequency components will lead to the loss of transient features, such as high-frequency oscillation information, and the positioning error will increase; if the threshold is too high (>0.35), too many noise-dominated intrinsic mode components will be retained, the feature signal-to-noise ratio will decrease, and the recognition accuracy will decrease.

[0169] 2. The impact of the distortion rate threshold on system performance is shown in Table 3.

[0170] Table 3

[0171]

[0172] As shown in Table 3, when the distortion rate is 5%, the transient waveform distortion rate (3.0%) and data storage volume (12.3 GB / hour) reach the optimal balance, resulting in the smallest positioning error (48 m). If the threshold is too low (<5%), the sampling rate frequently switches to 10 MHz, resulting in a slightly lower distortion rate (2.8%), but a surge in data volume (18.5 GB / hour), increasing hardware storage pressure. If the threshold is too high (>5%), the sampling rate switching lags, the transient signal is undersampled, the waveform distortion rate increases significantly, leading to a positioning error of 89 m.

[0173] Under high noise (signal-to-noise ratio 20dB) and large transition resistance (100Ω) fault conditions, the MPE threshold is in the range of 0.28~0.32, and the positioning error fluctuation is ≤±5m; the distortion rate threshold is in the range of 4%~6%, and the distortion rate fluctuation is ≤±0.5%.

[0174] For every deviation of 0.3±0.05 from the MPE threshold, the positioning error changes by 15% / 0.1% of the threshold; for every deviation of 5%±1% from the distortion rate threshold, the distortion rate changes by 2% / 1% of the threshold, indicating that the existing thresholds still maintain good performance under ±20% perturbation.

[0175] In summary, an MPE of 0.3 maximizes the preservation of effective transient characteristics, while a distortion rate of 5% balances transient waveform integrity with system resource consumption; these two values ​​represent the optimal range under current experimental conditions. The recommended MPE threshold range is 0.28–0.32, and the recommended distortion rate threshold range is 4%–6%. Exceeding these ranges will lead to a significant performance degradation.

[0176] Example 6

[0177] Embodiment 6 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the traveling wave model setting system and method based on waveform transient characteristics.

[0178] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0179] The memory can be used to store the computer program and / or modules. The processor implements various functions of the device for obtaining a leaderboard by running or executing the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart memory card, secure digital card, flash memory card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0180] Example 7

[0181] Embodiment 7 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the traveling wave model setting system and method based on waveform transient characteristics.

[0182] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0183] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0184] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for setting a traveling wave model based on waveform transient characteristics, characterized by, The method comprises the following steps: Step 1: decompose the second signal of the traveling wave sensor output into a plurality of intrinsic modal components by using complementary ensemble empirical mode decomposition; effectively screen the intrinsic modal components by using multi-sequence permutation entropy; Step 2: fit the waveform of the effective intrinsic modal component by using a nonlinear least squares optimization algorithm, wherein a parabolic function is used to fit the rising edge of the wave head, and a sinusoidal function is used to fit the oscillation decay section; Step 3: construct a fault feature feedback mechanism to obtain the waveform distortion rate and the frequency mutation range in real time; and adjust the sampling frequency according to the waveform distortion rate and the frequency mutation range; Step 4: extract time domain features and frequency domain features from the effective intrinsic modal component, wherein the time domain features include the amplitude mutation rate and the rise time, and the frequency domain features include the main frequency offset and the harmonic energy; and output the fault type and the position by using a deep forest model to weight and fuse the extracted time domain features and frequency domain features. In step 1, the effective intrinsic modal component is screened by using multi-sequence permutation entropy, and the screening process comprises the following steps:

2. The method of claim 1, wherein, Obtain the permutation entropy value of each intrinsic modal component; Determine the intrinsic modal component with a permutation entropy value lower than a preset entropy threshold value as the effective intrinsic modal component.

3. The traveling wave model setting method based on waveform transient characteristics according to claim 1, characterized in that, In step 2, the waveform of the effective intrinsic modal component is fitted by using a nonlinear least squares optimization algorithm, and the nonlinear least squares optimization algorithm determines the optimal fitting parameters by minimizing the residual sum of squares of the fitting curve and the real waveform, and the mathematical expression is: Wherein, yi is the sampling value of the effective IMF component, f(ti, β) is the combination function of the parabolic function and the sinusoidal function, β is the parameter vector to be optimized, and n is the number of sampling points. ； In step 3, the calculation formula of the waveform distortion rate δ is:

4. The method of claim 1, wherein, Wherein, Aactual is the actual amplitude of the signal in the current sampling period, and Atheoretical is the theoretical amplitude based on historical data or a preset model. ； The time domain features and frequency domain features extracted in step 4 include a total of 29 features.

5. The method of claim 1, wherein, In step 4, the deep forest model includes a plurality of random forests as base classifiers, and the 29 features are learned and fused layer by layer through a hierarchical structure to output the probability distribution of the fault type and the estimated value of the fault position.

6. The method of claim 5, wherein, When δ>5%, the sampling frequency is triggered to be increased.

7. The method of claim 4, wherein the method further comprises: The preset entropy threshold value MPEth=0.

3.

8. The method of claim 2, wherein the method further comprises: In step 4, the fault type and the position are output, and the output process comprises:

9. The method of claim 1, wherein, Output the fault position by dynamic weighting: Wherein, fault location is the estimated value of the fault position, wm is the weight of the mth base classifier, fm is the mth random forest classifier function, f is the 29-dimensional feature vector, and M is the total number of base classifiers. ； The traveling wave model setting system is used to execute the traveling wave model setting method according to any one of claims 1-9, and the traveling wave model setting system comprises a multi-modal signal decomposition and feature enhancement module, a wave head peak accurate capture module, a dynamic sampling frequency regulation module, and a multi-modal data fusion and fault classification module; 10. A traveling wave model setting system based on waveform transient characteristics, characterized by, The multi-modal signal decomposition and feature enhancement module comprises a decomposition unit and a screening unit; ​ The decomposition unit receives the secondary signals of the traveling wave sensor output and performs decomposition; the screening unit calculates the permutation entropy of each intrinsic modal component and screens the effective intrinsic modal component. The wave head peak accurate capture module includes a nonlinear least squares optimization algorithm optimizer and a wave head calculation unit. The nonlinear least squares optimization algorithm optimizer performs function fitting on the input effective intrinsic modal component; the wave head calculation unit determines the wave head peak point according to the fitting curve. The dynamic sampling frequency regulation module includes a signal analysis unit and a frequency control unit. The signal analysis unit is used to calculate the distortion rate and monitor the frequency mutation. The frequency control unit adjusts the sampling frequency according to the analysis result. The multi-modal data fusion and fault classification module includes a feature extraction sub-module and a deep forest processing sub-module. The feature extraction sub-module is used to extract 29-dimensional features. The deep forest processing sub-module is used to load the model and perform fusion calculation, and output the fault type and location.

Images