A Circuit Protection Method and Device Based on Current Data

Abstract

The present invention relates to the field of circuit protection analysis, and discloses a circuit protection method and device based on current data. The specific method steps include: performing hierarchical noise estimation on the original current signal to obtain hierarchical thresholds and dynamically iteratively denoising to avoid noise estimation deviation, dynamically optimizing the hierarchical thresholds by combining kurtosis and skewness, and making protection decisions through the analysis of the optimized denoised current signal; extracting and analyzing current features of the optimized denoised current signal, establishing and training a circuit fault model by using bidirectional LSTM combined with an attention mechanism, and based on the circuit fault model, correcting the current threshold in real time, comparing it with the output real-time current value of the optimized denoised current signal, and combining with current features for analysis to generate protection decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit protection analysis, and particularly to a circuit protection method and device based on current data. Background Art

[0002] In the field of circuit protection, the accurate analysis of current signals is crucial. With the increasing complexity of the power system and the continuous improvement of the requirements for circuit reliability, traditional circuit protection methods face many challenges. The original current signals are inevitably affected by various noise interferences during actual acquisition, making the effective processing of current signals a key link to ensure the accuracy of circuit protection. Most traditional signal denoising methods use global threshold processing. In a complex circuit environment, it is difficult to accurately capture the characteristics of noise in different frequency bands, and the global threshold cannot perform differential processing for each frequency band of noise. In terms of constructing a circuit fault model, early models lacked the full utilization of the timing information and key information of current signals. Since the occurrence and evolution of circuit faults are dynamic, it is difficult to accurately describe and predict the fault state relying on simple models. When traditional models are trained, they rely highly on data in specific scenarios and lack generality. When the application scenario changes, the performance and accuracy of the model will drop significantly.

[0003] In the prior art, most traditional denoising methods are based on the Gaussian noise assumption. The noise in the actual circuit contains a large number of outliers and off - points. The global threshold processing method relied on by traditional methods cannot adapt to the complex noise environment, is extremely sensitive to non - Gaussian noise and outliers, easily leads to large noise estimation deviations, and thus affects the denoising effect of the signal and subsequent analysis. The timing information in the current signal is not fully considered. The occurrence of a circuit fault is a dynamic process, and there is a correlation between the current signals at the front and back moments. Ignoring the timing information will cause the model to be unable to accurately capture the law of fault evolution and reduce the ability to predict and diagnose faults. Traditional models do not have an effective mechanism to highlight the key information in current signals. When dealing with large - scale data, it is difficult to focus on the data part that is of great significance for fault judgment, resulting in poor adaptability of the model to complex fault scenarios, being unable to accurately distinguish different types of faults, judge the severity of the faults, and make decisions in a timely manner. Traditional models rely heavily on a large amount of data in specific scenarios during training, and a large amount of data needs to be collected and labeled for training for different circuits and application scenarios, consuming a large amount of manpower, material resources and time. When facing new and under - trained scenarios, the generalization ability of the model is insufficient, making it difficult to accurately judge faults and unable to effectively protect the circuit in a timely manner.

[0004] In view of this, it is necessary to provide a circuit protection method and device based on current data. Summary of the Invention

[0005] The object of the present invention is to provide a circuit protection method and device based on current data. To solve the above-mentioned prior art problems, the present invention is achieved through the following technical solutions:

[0006] In the first aspect, an embodiment of the present invention provides a circuit protection method based on current data, which specifically includes the following steps:

[0007] Perform hierarchical noise estimation on the original current signal to obtain hierarchical thresholds and dynamically iterate for noise reduction to avoid noise estimation deviation;

[0008] Dynamically optimize the hierarchical thresholds by combining kurtosis and skewness, and make protection decisions through the analysis of the optimized noise-reduced current signal;

[0009] Extract and analyze the current characteristics of the optimized noise-reduced current signal, and establish and train a circuit fault model by using bidirectional LSTM combined with an attention mechanism;

[0010] Based on the circuit fault model, the current threshold is corrected in real time, compared with the real-time current value output by the optimized noise-reduced current signal, and combined with the current characteristics for analysis to generate a protection decision.

[0011] In the second aspect, an embodiment of the present invention provides a circuit protection system based on current data, which specifically includes the following units:

[0012] Signal preprocessing unit: Perform hierarchical decomposition on the original current signal by using wavelet transform, calculate the high-frequency coefficients of each layer and analyze to obtain hierarchical thresholds, perform soft threshold processing on the high-frequency coefficients of each layer to mark and exclude outliers, and perform secondary noise estimation and threshold correction through signal residuals for dynamic iterative noise reduction; Calculate the kurtosis and skewness of the current data of each layer based on the high-frequency coefficients, compare with the preset standard values to judge whether there are non-Gaussian noises and outliers, calculate the dynamically optimized hierarchical thresholds and perform dynamic iterative noise reduction accordingly to obtain the optimized noise-reduced current signal, and output the real-time current value;

[0013] Feature extraction unit: Extract time-domain, frequency-domain, time-frequency domain and non-linear features to generate a multi-dimensional feature vector;

[0014] Fault model unit: Construct a circuit fault model by using bidirectional LSTM combined with an attention mechanism, and combine Kirchhoff's law constraints to output the fault probability;

[0015] Protection decision unit: Based on the fault output probability output by the circuit fault model, use a preset formula to correct the current threshold in real time, combine the real-time current value output by the optimized noise-reduced current signal with the corrected current threshold, and comprehensively analyze with reference to the multi-dimensional features of the circuit current to generate a protection decision.

[0016] Advantages of the present invention:

[0017] 1. Multi - resolution decomposition is carried out using wavelet transform to achieve independent noise estimation of different frequency components. Compared with the traditional global threshold method, hierarchical processing can capture the noise characteristics of different frequency bands more accurately. The standard deviation is estimated based on the median absolute deviation and corrected through conversion coefficients to improve the robustness to non - Gaussian noise and outliers, avoiding the sensitivity problem of traditional methods to outliers. Through residual signal reconstruction and secondary noise estimation, iterative correction of the threshold is realized. It solves the problem of initial noise estimation deviation in traditional single - stage noise reduction, is applicable to non - stationary noise environments, and introduces kurtosis and skewness as statistical criteria to break through the limitation of traditional noise reduction methods relying on the Gaussian noise assumption. When non - Gaussian noise or outliers are detected, the threshold is adaptively increased by adjusting the coefficients to achieve a dynamic balance between noise suppression and signal fidelity.

[0018] 2. By extracting current characteristics, the characteristics of current signals are comprehensively described from different angles, accurately reflecting the operating state of the circuit. The circuit fault model combining bidirectional LSTM and attention mechanism fully utilizes the temporal information and key information of current signals, is pre - trained on a large - scale IEEE power fault dataset, enabling the model to learn general power fault patterns and characteristics, and is adapted to specific circuit fault protection scenarios through fine - tuning, reducing the dependence on a large amount of specific - scenario data. The method of real - time correcting the current threshold can dynamically adjust the current threshold according to the fault probability of the circuit, and generates protection decisions through comprehensive analysis of the real - time current value, corrected current threshold, and current characteristics, more comprehensively evaluating the operating state of the circuit and protecting the circuit. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the following - described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 is a flowchart of the steps of a circuit protection method based on current data provided in Embodiment 1 and Embodiment 2 of the present invention;

[0021] Figure 2 is a schematic structural diagram of a circuit protection system based on current data provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0023] Embodiment 1

[0024] A circuit protection method based on current data provided by an embodiment of the present invention, as Figure 1 shown, specifically includes the following steps:

[0025] Step 1: Perform hierarchical noise estimation on the original current signal to obtain hierarchical thresholds and dynamically iterate for noise reduction to avoid noise estimation deviation;

[0026] In a specific embodiment, the specific method for performing hierarchical noise estimation on the original current signal is as follows:

[0027] Use wavelet transform to perform a -layer decomposition transformation on the original current signal to obtain the high-frequency coefficients of the -layer, where is the number of layers into which the original current signal is decomposed and transformed;

[0028] For each layer of high-frequency coefficients , calculate the median of the absolute value of the high-frequency coefficient . Obtain the median absolute deviation through the formula . Divide the obtained median absolute deviation by the conversion coefficient to obtain the estimated standard deviation of the noise of the -layer;

[0029] It should be noted that the conversion coefficient represents the conversion relationship coefficient between the median and the standard deviation of the high-frequency coefficients, and the preset value is 0.6745. For a standard normal distribution, 67.45% of the data is within ±0.6745 times the standard deviation of the mean;

[0030] Based on the obtained estimated standard deviation , calculate the threshold for each layer of noise through the formula to obtain the hierarchical threshold , where represents the length of the high-frequency coefficients of the -layer;

[0031] In a specific embodiment, the specific method for dynamic iterative noise reduction to avoid noise estimation deviation is as follows:

[0032] Apply a hierarchical threshold to each layer of high-frequency coefficients , perform soft threshold processing, and obtain the iterative processing coefficient through the formula Calculate the iterative processing coefficient , analyze and judge based on the obtained processing coefficient, and mark the outliers of the high-frequency coefficients;

[0033] Specifically, if , it means that there are outliers in the high-frequency coefficients, and the high-frequency coefficient is marked as an outlier, and the high-frequency coefficients marked as outliers are excluded to obtain the reconstructed preliminary noise-reduced signal ;

[0034] Based on the reconstructed preliminary noise-reduced signal , subtract the original current signal from the preliminary noise-reduced signal to obtain the signal residual ;

[0035] Perform SWT decomposition on the obtained signal residual to obtain the residual high-frequency coefficients ;

[0036] For each layer of residual high-frequency coefficients , calculate the median of the absolute value of the residual high-frequency coefficient , and obtain the first corrected median absolute deviation through the formula , divide the obtained median absolute deviation by the conversion coefficient to obtain the first corrected estimated standard deviation of the noise of the th layer ;

[0037] Based on the obtained first corrected estimated standard deviation , calculate the threshold of the noise of each layer through the formula to obtain the first corrected hierarchical threshold , where represents the length of the high-frequency coefficients of the th layer;

[0038] Perform soft threshold processing on the obtained first corrected hierarchical threshold to obtain the second iterative processing coefficient , and perform dynamic iterative noise reduction on the original current signal based on the obtained iterative processing coefficient to obtain the th iterative processing coefficient , where is the maximum number of iterations;

[0039] ​Step 2: Dynamically optimize the hierarchical threshold by combining kurtosis and skewness, and make a protection decision through the analysis of the optimized noise-reduced current signal;

[0040] In a specific embodiment, the specific method for dynamically optimizing the hierarchical threshold by combining kurtosis and skewness is as follows:

[0041] Calculate the kurtosis of the current data of each layer based on the obtained high-frequency coefficients through the formula to calculate the data kurtosis , calculate the skewness of the current data of each layer based on the obtained high-frequency coefficients through the formula to calculate the data skewness , where represents the mean of the high-frequency coefficients of the th layer, and represents the standard deviation of the high-frequency coefficients of the th layer;

[0042] Compare the calculated kurtosis with the preset kurtosis standard value , and at the same time, compare the calculated skewness with the preset skewness standard value to analyze and judge whether non-Gaussian noise and outliers appear in the layering, and increase the hierarchical threshold;

[0043] Specifically, if it is determined that the kurtosis is greater than the preset kurtosis standard value and / or the skewness is greater than the preset skewness standard value, it indicates that there are non-Gaussian noise and outliers in the layering. Through the formula calculate the optimized hierarchical threshold after dynamic optimization , where represents the preset adjustment coefficient;

[0044] Perform dynamic iterative noise reduction based on the optimized hierarchical threshold after dynamic optimization to obtain the optimized noise-reduced current signal and output the real-time current value ;

[0045] Analyze the real-time current value output by the optimized noise-reduced current signal obtained through dynamic iterative noise reduction and the preset current threshold. When the real-time current value is greater than the preset current threshold, cut off the power supply of the circuit for protection;

[0046] The technical solution of the present invention is as follows: Wavelet transform is used for multi-resolution decomposition to achieve independent noise estimation of different frequency components. Compared with the traditional global threshold method, hierarchical processing can more accurately capture the noise characteristics of different frequency bands. Based on the median absolute deviation to estimate the standard deviation, through the conversion coefficient correction, the robustness to non-Gaussian noise and outliers is improved, avoiding the sensitivity problem of the traditional method to outliers. Through residual signal reconstruction and secondary noise estimation, the iteration correction of the threshold is realized. Solve the problem of initial noise estimation deviation in traditional single denoising, applicable to non-stationary noise environments. Introduce kurtosis and skewness as statistical criteria to break through the limitation of traditional denoising methods relying on the Gaussian noise assumption. When non-Gaussian noise or outliers are detected, adaptively increase the threshold by adjusting the coefficient to achieve the dynamic balance between noise suppression and signal fidelity.

[0047] Embodiment 2

[0048] As Figure 1 shown, a circuit protection method based on current data provided by an embodiment of the present invention specifically includes the following steps:

[0049] Step three: Extract and analyze the current characteristics of the optimized denoised current signal, and establish and train a circuit fault model by using bidirectional LSTM combined with an attention mechanism;

[0050] In a specific embodiment, the specific method for extracting the current characteristics of the optimized denoised current signal is:

[0051] Based on the optimized denoised current signal, obtain the current characteristics of the optimized denoised current signal, and the current characteristics include: time-domain characteristics, frequency-domain characteristics, time-frequency domain characteristics, and non-linear characteristics;

[0052] It should be noted that the time-domain characteristics of the optimized denoised current signal include but are not limited to: mean, variance, kurtosis, and waveform factor; the frequency-domain characteristics of the optimized denoised current signal include but are not limited to: total harmonic distortion rate, fundamental wave energy ratio, and specific frequency component ratio; the time-frequency domain characteristics of the optimized denoised current signal include but are not limited to: wavelet entropy and local mean decomposition energy distribution; the non-linear characteristics of the optimized denoised current signal include but are not limited to: permutation entropy and sample entropy;

[0053] Specifically, based on the optimized denoised current signal, set sampling points, and obtain current sampling values of sampling current signals, and calculate the mean of the sampling current signal based on the obtained current sampling values through the mean formula;

[0054] Calculate the variance of the sampling current signal by using the mean of the calculated sampling current signal and the current sampling values through the variance formula;

[0055] The mean value of the sampled current signal is calculated and used in the formula to calculate the form factor of the sampled current signal , where represents the number of sampled current signals, represents the th sampled current signal, represents the mean value of the sampled current signal;

[0056] The harmonic effective value of the current signal is obtained, and the total harmonic distortion rate of the current signal is obtained through the calculation formula , where represents the specific harmonic order, represents the th harmonic effective value, represents the effective value of the fundamental wave;

[0057] The energy of the fundamental wave of the current signal is divided by the total harmonic energy to obtain the proportion of the energy of the fundamental wave of the current signal in the total harmonic energy;

[0058] The proportion of the energy of a specific harmonic order is obtained. For the identification of specific circuit faults, the proportion of a specific frequency component is obtained through the calculation formula , where represents the specific harmonic order, represents the th harmonic energy;

[0059]

[0060] After the signal is decomposed by wavelet transform, the wavelet entropy is obtained by calculating the information entropy of the energy distribution in each frequency band, which measures the complexity of the signal;

[0061] The signal is decomposed into local characteristic scale components (Product Function components), and the proportion of the energy of each component is calculated to obtain the local mean decomposition energy distribution, which is used to analyze the transient components in the non-stationary signal;

[0062] It should be noted that the local characteristic scale component represents a representation form for describing the local characteristics of signal data at different scales;

[0062] The permutation entropy is calculated through the permutation pattern of the time series. The larger the permutation entropy, the stronger the randomness;

[0063] The sample entropy of the current signal is obtained, which measures the probability of pattern repetition in the time series. The smaller the sample entropy, the stronger the regularity;

[0064] In a specific embodiment, the specific method for establishing and training a circuit fault model by using a bidirectional LSTM combined with an attention mechanism is as follows: [[ID=②]]

[0065] A circuit fault model is established by combining a Bidirectional Long Short-Term Memory Network (Bidirectional LSTM) with an attention mechanism;

[0066] The circuit fault model includes: a current feature input layer, a Bidirectional LSTM layer, an attention mechanism layer, and a fully connected layer;

[0067] It should be noted that the Bidirectional Long Short-Term Memory Network is an extension based on the Long Short-Term Memory Network LSTM and consists of two LSTMs with opposite directions. One forward LSTM processes the input data from the beginning to the end of the sequence, and the other backward LSTM processes the input data from the end to the beginning of the sequence. The Bidirectional LSTM captures the forward and backward temporal information and processes the dynamic characteristics of the circuit fault evolution. The attention mechanism is a technology widely used in the fields of deep learning and artificial intelligence, which simulates the way of attention allocation in the human visual system, enables the model to focus on the important information in the input data, ignores the irrelevant or secondary information, thereby improving the performance and efficiency of the model, dynamically focuses on the key parts of the input sequence, and improves the performance of the model;

[0068] The multi-dimensional current features obtained through calculation are input into the Bidirectional LSTM layer. Kirchhoff's law of circuit equations is introduced as a regularization term in the Bidirectional LSTM to constrain the model to learn features that conform to the physical laws of the circuit, and the forward and backward input current feature sequences are processed through the Bidirectional LSTM layer;

[0069] The output of the Bidirectional LSTM layer is weighted by the attention mechanism, and the output of the attention mechanism layer is mapped to the output of the fault probability to obtain the fault output probability ;

[0070] Specifically, the current circuit fault type is mapped to a query vector through the fully connected layer to obtain a query vector , for the hidden state at each time step A linear transformation is performed through the formula to obtain the key vector , where represents the linear transformation weight matrix that maps the hidden state to the key vector space, which determines the contribution weight of different features to the attention calculation, represents adding a learnable offset to the key vector;

[0071] Weight normalization is performed through the attention score formula to obtain the importance weight of the time step for fault determination ;

[0072] The hidden state A linear transformation is performed through the formula to calculate the value vector , weighted fusion to obtain a concentrated vector ;

[0073] Through the fully connected layer, after non-linear transformation map to output the fault output probability , where represents the weight matrix of the fully connected layer, represents the offset of the fully connected layer, represents the weight matrix of the fault output probability, represents the offset of the fault output probability, the fault output probability ;

[0074] Pre-train on the IEEE power fault dataset, fine-tune the pre-trained model to the circuit fault protection scenario, use the training dataset to train the model, and adopt the L2 regularization loss function to prevent overfitting;

[0075] Step 4: Based on the circuit fault model, compare the real-time corrected current threshold with the optimized noise-reduced current signal output real-time current value and analyze in combination with current characteristics to generate a protection decision;

[0076] In a specific embodiment, the specific method for making a protection decision on the circuit based on the circuit fault model by real-time correcting the current threshold and analyzing in combination with current characteristics is:

[0077] Perform real-time correction of the current threshold on the calculated fault output probability, and obtain the corrected current threshold through the formula where , where represents the preset current threshold, represents the preset sensitivity coefficient, with a value of 0.3;

[0078] Compare the real-time current value output by the obtained optimized noise-reduced current signal with the corrected current threshold and analyze in combination with current characteristics to generate a protection decision;

[0079] Exemplarily, set the circuit preset current threshold to 10A, and the sensitivity coefficient to 0.3; when the fault output probability output by the model is 0.1, the corrected current threshold calculated through the formula is 9.7A, and the real-time current value output by the optimized noise-reduced current signal is 8A. Through analysis, it is obtained that , and at the same time, combining the analysis that the current characteristics are all within the normal range, it is judged that the circuit is in normal operation and no protection operation is required; when the fault output probability is 0.5, and the corrected current threshold is obtained through formula calculation is 8.5A, optimizing the real-time current value of the noise-reduced current signal output is 8.2A, obtained through analysis , and at the same time, by analyzing that the total harmonic distortion rate in the current characteristics suddenly increases, there is a potential fault risk in the circuit, an early warning signal is sent to remind the operation and maintenance personnel to check; when the fault output probability output by the model is 0.9, and the corrected current threshold is obtained through formula calculation is 7.3A, optimizing the real-time current value of the noise-reduced current signal output is 8A, obtained through analysis , and at the same time, by analyzing that the waveform factor in the current characteristics is abnormal, it is judged that the circuit has a fault, and the circuit is immediately powered off to prevent the fault from further expanding;

[0080] The technical solution of the embodiment of the present invention is as follows: based on the optimized noise-reduced current signal, current characteristics in the time domain, frequency domain, time-frequency domain and non-linearity are obtained, and a circuit fault model is established by combining bidirectional LSTM with an attention mechanism. The bidirectional LSTM consists of two LSTMs with opposite directions, capturing forward and backward timing information. Kirchhoff's law of circuit equations is introduced as a regularization term in the bidirectional LSTM to constrain the model to learn features that conform to the physical laws of the circuit and follow the basic physical principles of the circuit. The attention mechanism performs attention weighting on the output of the bidirectional LSTM layer, dynamically focusing on the key parts in the input sequence, enabling the model to focus on important information and ignore irrelevant or secondary information. The model is pre-trained on the IEEE power fault dataset and fine-tuned to the circuit fault protection scenario, trained using the training dataset, and the L2 regularization loss function is used to prevent overfitting, improving the generalization ability and stability of the model. The calculated fault output probability is used to correct the current threshold in real time through a specific formula. The real-time current value of the optimized noise-reduced current signal output is combined with the current characteristics of the corrected current threshold for analysis, and corresponding protection decisions are generated according to different situations; by extracting current characteristics, the characteristics of the current signal are comprehensively described from different angles, accurately reflecting the operating state of the circuit. The circuit fault model combining bidirectional LSTM with an attention mechanism makes full use of the timing information and key information of the current signal, is pre-trained on the large-scale IEEE power fault dataset, enables the model to learn general power fault patterns and characteristics, adapts to specific circuit fault protection scenarios through fine-tuning, reduces the dependence on a large amount of specific scenario data, and the method of correcting the current threshold in real time can dynamically adjust the current threshold according to the fault probability of the circuit, and combines the real-time current value, corrected current threshold and current characteristics for comprehensive analysis to generate protection decisions, more comprehensively evaluating the operating state of the circuit and protecting the circuit.

[0081] Embodiment III

[0082] As Figure 2 shown, a circuit protection system based on current data provided by an embodiment of the present invention specifically includes the following units:

[0083] Signal preprocessing unit: Perform hierarchical decomposition on the original current signal using wavelet transform, calculate the high-frequency coefficients of each layer and analyze to obtain the hierarchical threshold, perform soft threshold processing on the high-frequency coefficients of each layer to mark and exclude outliers, and perform secondary noise estimation and threshold correction through signal residuals for dynamic iterative noise reduction; Calculate the kurtosis and skewness of the current data of each layer based on the high-frequency coefficients, compare with the preset standard value to determine whether there are non-Gaussian noises and outliers, calculate the dynamically optimized hierarchical threshold and perform dynamic iterative noise reduction based on this to obtain an optimized noise-reduced current signal, and output the real-time current value;

[0084] Feature extraction unit: Extract time-domain, frequency-domain, time-frequency domain, and non-linear features to generate a multi-dimensional feature vector;

[0085] Fault model unit: Construct a circuit fault model using bidirectional LSTM combined with an attention mechanism, and combine Kirchhoff's law constraints to output the fault probability;

[0086] Protection decision unit: Based on the fault output probability output by the circuit fault model, use a preset formula to correct the current threshold in real time, combine the real-time current value output by the optimized noise-reduced current signal with the corrected current threshold, and simultaneously refer to the multi-dimensional features of the circuit current for comprehensive analysis to generate a protection decision.

[0087] The above has described an embodiment of the present invention in detail, but the content described is only the preferred embodiment of the present invention and cannot be considered as limiting the implementation scope of the present invention; The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by software simulation of a large amount of collected data to obtain a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation and historical experience and can be adjusted according to the actual situation; The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any equal changes and improvements made according to the scope of the present invention application should still fall within the patent coverage scope of the present invention.

Claims

1. A circuit protection method based on current data, characterized in that It includes the following steps: Perform hierarchical noise estimation on the original current signal to obtain hierarchical thresholds and dynamically iterate for noise reduction to avoid noise estimation bias; Dynamically optimize the hierarchical thresholds by combining kurtosis and skewness, and make protection decisions through the analysis of the optimized noise-reduced current signal; Extract and analyze the current characteristics of the optimized noise-reduced current signal, and establish and train a circuit fault model using bidirectional LSTM combined with an attention mechanism; Based on the circuit fault model, the current threshold is corrected in real time and compared with the optimized noise-reduced current signal output real-time current value, and combined with the current characteristics for analysis to generate a protection decision.

2. The circuit protection method based on current data according to claim 1, characterized in that, The method for performing hierarchical noise estimation on the original current signal is as follows: The original current signal is subjected to wavelet transform to obtain the high-frequency coefficients of each layer through hierarchical decomposition transform ; For each layer of high-frequency coefficients, calculate the median of the absolute values of the high-frequency coefficients, calculate the median absolute deviation, and divide the obtained median absolute deviation by the conversion coefficient to obtain the estimated standard deviation of the noise.

3. The circuit protection method based on current data according to claim 1, wherein, The method for obtaining the hierarchical threshold is as follows: Based on the obtained estimated standard deviation Calculate the hierarchical threshold for the noise of each layer , where represents the length of the high-frequency coefficients of the 4. A circuit protection method based on current data according to claim 1, characterized in that, The specific method for dynamic iterative noise reduction is as follows: Apply hierarchical thresholds to the high-frequency coefficients of each layer, perform soft thresholding, and obtain the iterative processing coefficients through formula calculation. , based on the obtained processing coefficients, perform analysis and judgment, mark the outliers of the high-frequency coefficients, and subtract the original current signal from the preliminary noise-reduced signal to obtain the signal residual. Perform SWT decomposition on the obtained signal residual to obtain the residual high-frequency coefficients ; For each layer of residual high-frequency coefficients , calculate the median of the absolute values of the residual high-frequency coefficients , and obtain the first corrected median absolute deviation . Divide the obtained median absolute deviation by the conversion coefficient to obtain the first corrected estimated standard deviation of the noise of the th layer ; Based on the obtained first corrected estimated standard deviation The noise of each layer is used to calculate the first corrected layer threshold , where represents the length of the high-frequency coefficients of the The obtained first corrected layering threshold Perform soft threshold processing to obtain the second iteration processing coefficient , and perform dynamic iterative noise reduction on the original current signal based on the obtained iteration processing coefficient to obtain the th iteration processing coefficient , where is the maximum number of iterations.

5. A circuit protection method based on current data according to claim 1, characterized in that, The method for dynamically optimizing the hierarchical threshold by combining kurtosis and skewness is as follows: Calculate the kurtosis of the current data for each layer based on the obtained high-frequency coefficients, and obtain the data kurtosis through formula calculation , calculate the skewness of the current data for each layer based on the obtained high-frequency coefficients, and obtain the data skewness through formula calculation ; If it is determined that the kurtosis is greater than the preset kurtosis standard value or the skewness is greater than the preset skewness standard value, it indicates that there are non-Gaussian noises and outliers in the layering, and the optimized layering threshold after dynamic optimization is calculated through the formula .

6. The circuit protection method based on current data according to claim 1, wherein The method for establishing a circuit fault model using bidirectional LSTM combined with an attention mechanism is as follows: Input the multi-dimensional current characteristics obtained by calculation into the bidirectional LSTM layer, introduce Kirchhoff's law of circuit equations as a regularization term in the bidirectional LSTM to constrain the model to learn features that conform to circuit physical laws, and process the forward and reverse input current feature sequences through the bidirectional LSTM layer; The output of the bidirectional LSTM layer is weighted by attention through the attention mechanism, and the output of the attention mechanism layer is mapped to the output of the fault probability to obtain the fault output probability .

7. A circuit protection method based on current data according to claim 6, characterized in that, The specific method for obtaining the fault output probability is as follows: Map the current circuit fault type to a query vector through a fully connected layer to obtain the query vector , for the hidden state at each time step The key vector is obtained through a linear transformation by the formula ; The hidden state linear transformation calculates the value vector through a formula , and the weighted fusion obtains the concentrated vector ; Output the fault output probability through the fully connected layer after non - linear transformation mapping , the fault output probability .

8. A circuit protection method based on current data according to claim 1, wherein The method for model training is as follows: Perform pre-training on the IEEE power fault data set, fine-tune the pre-trained model to the circuit fault protection scenario, use the training data set to train the model, and adopt the L2 regularization loss function.

9. A circuit protection method based on current data according to claim 1, characterized in that The method for the real-time correction of the current threshold and the comparison of the optimized noise-reduced current signal output real-time current value and the combination of current characteristics for analysis to generate a protection decision is as follows: The calculated fault output probability is used to correct the current threshold in real time through the formula to obtain the corrected current threshold , where represents the preset current threshold represents the preset sensitivity coefficient, and its value is 0.3; Analyze the real-time current value output by the optimized noise-reduced current signal obtained and the corrected current threshold in combination with the current characteristics to generate a protection decision.

10. A circuit protection system based on current data, which is used to execute the circuit protection method described in any one of the above claims 1-9, characterized in that, It includes the following units: Signal preprocessing unit: Perform hierarchical noise estimation on the original current signal to obtain hierarchical thresholds and dynamically iterate for noise reduction to avoid noise estimation bias, dynamically optimize the hierarchical thresholds by combining kurtosis and skewness, and obtain the optimized noise-reduced current signal; Feature extraction unit: Extract and analyze the current characteristics of the optimized noise-reduced current signal; Fault model unit: Establish and train a circuit fault model using bidirectional LSTM combined with an attention mechanism; Protection decision unit: Based on the circuit fault model, the current threshold is corrected in real time and compared with the optimized noise-reduced current signal output real-time current value, and combined with the current characteristics for analysis to generate a protection decision.

Images