Surge protection device nanosecond pulse test data processing method and system
By preprocessing the training pulse parameters and constructing a feature prediction model, the problem of complex interference in nanosecond pulse testing was solved, and a more accurate performance evaluation of surge protection devices was achieved.
Patent Information
- Application Number
- CN202511888909.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-02-06
AI Technical Summary
Existing data processing methods cannot effectively eliminate complex interference types in nanosecond pulse tests, leading to inaccurate performance evaluation of surge protection devices.
By acquiring training pulse parameters for data preprocessing, a target feature prediction model is constructed, multi-scale time-series feature sequence analysis is performed, and failure precursors are judged by combining historical data.
This improves the accuracy of surge protection device performance evaluation and avoids the bias caused by single-dimensional features or simple models.
Smart Images
Figure CN121476802A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nanosecond pulse test data processing technology for surge protection devices, and in particular to a method and system for processing nanosecond pulse test data for surge protection devices. Background Technology
[0002] Surge protection devices, as core components for suppressing surge energy and ensuring system safety, are widely used in critical fields such as communication base stations, power substations, rail transportation, and aerospace. Their performance reliability directly determines the anti-interference capability and operational safety of the entire system. Therefore, conducting device performance testing through nanosecond pulse injection experiments has become an essential step in the research, development, production inspection, and field operation and maintenance of surge protection devices. Precise and efficient test data processing technology is the core foundation for ensuring the validity of test results and supporting device performance evaluation.
[0003] Currently, existing data processing methods mostly employ simple noise filtering or baseline correction techniques, which cannot adequately address the complex interference types in nanosecond pulse testing. They are also unable to effectively eliminate the effects of electromagnetic coupling, cable crosstalk, and other factors, resulting in significant errors in the processed data. Consequently, the data cannot accurately reflect the true operating status of surge protection devices, thus reducing the accuracy of surge protection device performance evaluation. Summary of the Invention
[0004] This invention provides a method and system for processing nanosecond pulse test data of surge protection devices. It solves the technical problem that existing data processing methods mostly use simple noise filtering or baseline correction methods, which cannot fully cope with the complex interference types in nanosecond pulse testing, and cannot effectively eliminate the influence of factors such as electromagnetic coupling and cable crosstalk. As a result, the processed data still has large errors and cannot accurately reflect the true working state of surge protection devices, thus reducing the accuracy of surge protection device performance evaluation.
[0005] The first aspect of this invention provides a method for processing nanosecond pulse test data of surge protection devices, comprising:
[0006] Multiple training pulse parameters are acquired, and data preprocessing is performed on each training pulse parameter to obtain the corresponding time series feature set;
[0007] The preset initial feature prediction model is trained using the time-series feature set to obtain the corresponding target feature prediction model;
[0008] The nanosecond pulse test data of the surge protection device is acquired, and the target feature prediction model is used to predict the features of the nanosecond pulse test data to obtain the corresponding multi-scale time series feature sequence.
[0009] Based on the pre-acquired historical multi-scale time series feature sequence set, failure precursor analysis is performed on the multi-scale time series feature sequence to obtain the corresponding analysis results.
[0010] Optionally, the step of training a preset initial feature prediction model using the temporal feature set to obtain a corresponding target feature prediction model includes:
[0011] The time-series feature set is input into a preset initial feature prediction model for training to obtain the corresponding training feature sequence data;
[0012] Based on a preset loss function, the loss function value of the time-series feature set is calculated according to the training feature sequence data;
[0013] When the loss function value is greater than or equal to a preset loss threshold, the network parameters of the initial feature prediction model are adjusted until the loss function value is less than the loss threshold.
[0014] When the loss function value is less than the loss threshold, a target feature prediction model is generated.
[0015] Optionally, the target feature prediction model includes a preprocessing network, a feature decoupling network, and a temporal attention fusion layer. The step of performing feature prediction on the nanosecond pulse test data using the target feature prediction model to obtain the corresponding multi-scale temporal feature sequence includes:
[0016] The nanosecond pulse test data is preprocessed by the preprocessing network to obtain the corresponding data feature vector. The preprocessing network includes a first fully connected layer and a second fully connected layer connected in sequence.
[0017] The feature decoupling network is used to perform multi-scale feature extraction on the data feature vector to obtain time-domain feature vector, frequency-domain feature vector and time-frequency-domain feature vector;
[0018] The temporal attention fusion layer fuses the temporal feature vector, the frequency feature vector, and the time-frequency domain feature vector to obtain the corresponding multi-scale temporal feature sequence.
[0019] Optionally, the feature decoupling network includes a time-domain feature branch, a frequency-domain feature branch, and a time-frequency-domain feature branch. The step of performing multi-scale feature extraction on the data feature vector through the feature decoupling network to obtain the time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector includes:
[0020] Temporal features are extracted from the data feature vector through the temporal feature branch to obtain the corresponding temporal feature vector. The temporal feature branch includes a first fully connected layer and a bidirectional long short-term memory network connected in sequence.
[0021] The frequency domain feature vector is obtained by extracting frequency domain features from the data feature vector through the frequency domain feature branch. The frequency domain feature branch includes a second fully connected layer, a one-dimensional convolutional layer, a global average pooling layer and a third fully connected layer connected in sequence.
[0022] The time-frequency domain feature vector is extracted by the time-frequency domain feature branch to obtain the corresponding time-frequency domain feature vector. The time-frequency domain feature branch includes a first fully connected layer, a wavelet packet transform layer, an attention weight layer and a third fully connected layer connected in sequence.
[0023] Optionally, the step of performing failure precursor analysis on the multi-scale time-series feature sequences based on a pre-acquired set of historical multi-scale time-series feature sequences to obtain corresponding analysis results includes:
[0024] The least squares method is used to perform linear fitting on the pre-acquired historical multi-scale time series feature sequence set and the multi-scale time series feature sequence to obtain the corresponding trend term slope.
[0025] The multi-scale time-series feature sequence is compared with the corresponding failure precursor judgment threshold to obtain the corresponding comparison result;
[0026] When the comparison result indicates that the device has signs of impending failure, it is determined whether the slope of the trend term is greater than a preset slope threshold.
[0027] When the slope of the trend term is greater than the slope threshold, the analysis result is determined to be a device anomaly.
[0028] When the slope of the trend term is less than or equal to the slope threshold, the analysis result is determined to be that the device is normal.
[0029] When the comparison result indicates that the device is normal, the analysis result is determined to be that the device is normal.
[0030] Optionally, the configuration process of the slope threshold is as follows:
[0031] Obtain the slopes of multiple standard trend terms, and average the slopes of each standard trend term to obtain the corresponding first mean.
[0032] The preset safety margin coefficient is multiplied by the standard deviation of the slope of each of the standard trend terms to obtain the corresponding first multiplication value.
[0033] The first mean and the first multiplier are summed to obtain the corresponding slope threshold.
[0034] A second aspect of the present invention provides a nanosecond pulse test data processing system for surge protection devices, comprising:
[0035] The preprocessing module is used to acquire multiple training pulse parameters, perform data preprocessing on each training pulse parameter, and obtain the corresponding time series feature set;
[0036] The training module is used to train the preset initial feature prediction model using the time-series feature set to obtain the corresponding target feature prediction model.
[0037] The prediction module is used to acquire nanosecond pulse test data of surge protection devices, and to perform feature prediction on the nanosecond pulse test data through the target feature prediction model to obtain the corresponding multi-scale time series feature sequence.
[0038] The failure precursor analysis module is used to perform failure precursor analysis on the multi-scale time series feature sequences based on a pre-acquired set of historical multi-scale time series feature sequences, and obtain the corresponding analysis results.
[0039] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the surge protection device nanosecond pulse test data processing method as described in any of the preceding claims.
[0040] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the surge protection device nanosecond pulse test data processing method as described in any of the preceding claims.
[0041] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the surge protection device nanosecond pulse test data processing method as described in any of the preceding claims.
[0042] As can be seen from the above technical solutions, the present invention has the following advantages:
[0043] This invention acquires multiple training pulse parameters, preprocesses each parameter to obtain a corresponding time-series feature set, trains a pre-defined initial feature prediction model using this feature set to obtain a corresponding target feature prediction model, acquires nanosecond pulse test data of surge protection devices, and uses the target feature prediction model to predict features from this data, obtaining a corresponding multi-scale time-series feature sequence. Based on a pre-acquired set of historical multi-scale time-series feature sequences, it performs failure precursor analysis on the multi-scale time-series feature sequence to obtain the corresponding analysis results. This overcomes the technical problem that existing data processing methods often rely on simple noise filtering or baseline correction, which cannot accurately reflect the true operating state of surge protection devices and reduce the accuracy of surge protection device performance evaluation. This invention preprocesses the parameters of each training pulse to obtain the corresponding time-series feature set, providing a high-quality time-series feature set for the training of the subsequent initial feature prediction model. Then, the target feature prediction model is used to predict the features of the nanosecond pulse test data to obtain the corresponding multi-scale time-series feature sequence. Based on the pre-acquired historical multi-scale time-series feature sequence set, the multi-scale time-series feature sequence is analyzed for failure precursors to obtain the corresponding analysis results. This avoids the one-sidedness of evaluation caused by single-dimensional features or simple models and improves the accuracy of surge protection device performance evaluation. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart of the steps in a method for processing nanosecond pulse test data of a surge protection device according to Embodiment 1 of the present invention;
[0046] Figure 2 This is a flowchart illustrating the steps of a surge protection device nanosecond pulse test data processing method according to Embodiment 2 of the present invention.
[0047] Figure 3 This is a schematic diagram of the target feature prediction model provided in Embodiment 2 of the present invention;
[0048] Figure 4 The voltage waveform diagram for the experiment provided in Embodiment 2 of the present invention;
[0049] Figure 5 This is a schematic diagram of the setup for the nanosecond pulse injection experiment provided in Embodiment 2 of the present invention;
[0050] Figure 6 This is a structural block diagram of a surge protection device nanosecond pulse test data processing system provided in Embodiment 3 of the present invention;
[0051] Figure 7 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation
[0052] This invention provides a method and system for processing nanosecond pulse test data of surge protection devices. It addresses the technical problem that existing data processing methods often employ simple noise filtering or baseline correction, which cannot adequately handle the complex interference types in nanosecond pulse testing. These methods struggle to effectively eliminate the influence of factors such as electromagnetic coupling and cable crosstalk, resulting in significant errors in the processed data. Consequently, the data fails to accurately reflect the true operating state of the surge protection device, thus reducing the accuracy of surge protection device performance evaluation.
[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.
[0054] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a method for processing nanosecond pulse test data for surge protection devices according to Embodiment 1 of the present invention.
[0055] This invention provides a method for processing nanosecond pulse test data of surge protection devices, comprising:
[0056] Step 101: Obtain multiple training pulse parameters, perform data preprocessing on each training pulse parameter, and obtain the corresponding time series feature set.
[0057] Training pulse parameters refer to the pulse parameters, device voltage, and original waveform data of the current in the second pulse injection experiment.
[0058] A time series feature set refers to a dataset consisting of "input-output" sample pairs. The input is the impulse parameters in the experiment, and the output is a multi-scale time series feature sequence obtained through preprocessing and feature extraction (the multi-scale time series feature sequence includes time domain features, frequency domain features, and time-frequency domain features).
[0059] In this embodiment of the invention, a second pulse injection experiment is conducted through a preset second pulse injection experimental platform to obtain multiple training pulse parameters. Data preprocessing (i.e., three-level preprocessing and multi-scale feature extraction) is performed on each training pulse parameter to obtain the corresponding time series feature set.
[0060] It should be noted that the three-level preprocessing specifically includes: A1. Adaptive wavelet thresholding denoising is performed on the original waveform data of device voltage and current in each training pulse parameter (using the db6 wavelet basis adapted to the double exponential waveform, determining the number of 5 decomposition layers according to the signal-to-noise ratio maximization criterion, and dynamically calculating the threshold of each layer through Stein unbiased risk estimation (SURE) (formula)). N=2000 is the number of sampling points. The standard deviation of noise. ≤0.05kV), after soft thresholding of wavelet coefficients, inverse transform is used to reconstruct the denoised waveform. A2, EMD decomposition is performed on each original voltage waveform data and each original current waveform data after adaptive wavelet threshold denoising to obtain multiple intrinsic mode functions (for example, EMD decomposition is performed on a current waveform data to obtain 8-12 intrinsic mode functions (IMF1-IMF12) and 1 residual component (Res)). The "instantaneous frequency" of each intrinsic mode function is calculated by Hilbert transform, and intrinsic mode functions with an "instantaneous frequency" less than 1MHz are removed to obtain multiple corrected original voltage waveform data and original current waveform data. A3, outlier removal is performed on each corrected original voltage waveform data and original current waveform data (i.e., based on the mean and standard deviation of each original voltage waveform data and each original current waveform data, outliers exceeding [] are removed from the original voltage waveform data and original current waveform data. -3 , +3 The points to be removed are: The mean, (where the standard deviation is 1), resulting in multiple preprocessed training pulse parameters.
[0061] It should be noted that the multi-scale feature extraction is specifically as follows: B1. Based on the preset time-domain feature extraction function, extract the time-domain features of each preprocessed training pulse parameter. The time-domain features include rising edge steepness, oscillation attenuation coefficient, current-voltage synchronization, stable segment fluctuation, pulse recovery time (i.e., the time it takes for the device voltage to drop from the peak value to the 10% overshoot voltage peak value) and accumulated energy.
[0062] The time-domain feature extraction function is as follows:
[0063]
[0064]
[0065]
[0066]
[0067]
[0068] in, For the steepness of the rising edge, This is the peak value of the overshoot voltage. This represents the time corresponding to 90% of the peak value. The time corresponding to 10% peak value. This is the oscillation damping coefficient. This is the absolute value of the first trough of the oscillation. For current and voltage synchronization, The peak current time of the device. For the peak voltage time of the device, For the volatility of the stable segment, The standard deviation of voltage within a stable voltage range, To stabilize the voltage, To accumulate energy, This is the data of the device voltage waveform at time t. This is the data of the device current waveform at time t. This is the time index, and us is the sampling duration.
[0069] B2. The raw voltage and current waveforms of each preprocessed training pulse parameter are converted to the frequency domain using Fast Fourier Transform (FFT) to extract multiple frequency domain features. These features include characteristic frequency (the frequency corresponding to the maximum peak value in the power spectrum, reflecting the device's parasitic inductance / capacitance), bandwidth (bandwidth (BW) is the frequency range of the maximum power (≥50%) in the power spectrum; a wider bandwidth allows for a wider range of pulse frequencies to be adapted to the device), and high-frequency energy ratio (the high-frequency energy ratio (η) is the ratio of power at frequencies ≥100MHz to the total power; a larger η indicates greater energy loss of the device under high-frequency pulses). B3. Wavelet packet decomposition (WPD) is used to perform time-frequency analysis on the raw voltage waveforms of each preprocessed training pulse parameter to extract multiple time-frequency domain features. Among them, the time-frequency domain features include wavelet packet energy entropy (wavelet packet energy entropy (H): decompose WPD into 8 frequency bands (0-1GHz, 125MHz per band), calculate the entropy value of the energy proportion of each frequency band. The smaller the entropy value, the more concentrated the energy distribution and the more stable the device performance) and rising edge dominant frequency band (the rising edge dominant frequency band refers to the frequency band with the largest energy during the rising edge period after WPD decomposition, reflecting the main frequency components of the rising edge).
[0070] Step 102: Train the preset initial feature prediction model using the temporal feature set to obtain the corresponding target feature prediction model.
[0071] In this embodiment of the invention, mean squared error (MSE) or mean absolute error (MAE) is used as the loss function, and a time series feature set is used to train a preset initial feature prediction model to obtain the corresponding target feature prediction model.
[0072] It is worth mentioning that the Random Forest (RF) algorithm can be used to construct the initial feature prediction model. Integrating multiple decision trees using the Random Forest (RF) algorithm can improve the accuracy and stability of predictions.
[0073] It is worth mentioning that when training the initial feature prediction model, PLSR can be used to perform principal component analysis on the time series feature set to obtain the dimensionality-reduced time series feature set.
[0074] Step 103: Obtain the nanosecond pulse test data of the surge protection device, and perform feature prediction on the nanosecond pulse test data through the target feature prediction model to obtain the corresponding multi-scale time series feature sequence.
[0075] Nanosecond pulse test data refers to the electrical parameter data of surge protection devices collected through nanosecond pulse injection experiments, including voltage waveform data across the surge protection device and current waveform data in the device circuit.
[0076] In this embodiment of the invention, the nanosecond pulse test data of the surge protection device within the current sampling period is obtained, and the nanosecond pulse test data is used to predict the features of the target feature prediction model to obtain the corresponding multi-scale time series feature sequence.
[0077] Step 104: Perform failure precursor analysis on the multi-scale time series feature sequences based on the pre-acquired historical multi-scale time series feature sequence set, and obtain the corresponding analysis results.
[0078] In this embodiment of the invention, the least squares method is used to linearly fit the pre-acquired historical multi-scale time-series feature sequence set and the multi-scale time-series feature sequence to obtain the corresponding trend term slope. The multi-scale time-series feature sequence is compared with the corresponding failure precursor judgment threshold to obtain the corresponding comparison result. When the comparison result indicates that the device has a failure precursor, it is determined whether the trend term slope is greater than the preset slope threshold. When the trend term slope is greater than the slope threshold, the analysis result is determined to be device abnormal. When the trend term slope is less than or equal to the slope threshold, the analysis result is determined to be device normal. When the comparison result indicates that the device is normal, the analysis result is determined to be device normal.
[0079] In this embodiment of the invention, multiple training pulse parameters are acquired, and data preprocessing is performed on each training pulse parameter to obtain a corresponding time-series feature set. The time-series feature set is then used to train a preset initial feature prediction model to obtain a corresponding target feature prediction model. Nanosecond pulse test data of the surge protection device is acquired, and feature prediction is performed on the nanosecond pulse test data using the target feature prediction model to obtain a corresponding multi-scale time-series feature sequence. Based on a pre-acquired set of historical multi-scale time-series feature sequences, failure precursor analysis is performed on the multi-scale time-series feature sequence to obtain the corresponding analysis results. This overcomes the technical problem that existing data processing methods often employ simple noise filtering or baseline correction methods, which cannot accurately reflect the true working state of the surge protection device and reduce the accuracy of surge protection device performance evaluation. This invention preprocesses the parameters of each training pulse to obtain the corresponding time-series feature set, providing a high-quality time-series feature set for the training of the subsequent initial feature prediction model. Then, the target feature prediction model is used to predict the features of the nanosecond pulse test data to obtain the corresponding multi-scale time-series feature sequence. Based on the pre-acquired historical multi-scale time-series feature sequence set, the multi-scale time-series feature sequence is analyzed for failure precursors to obtain the corresponding analysis results. This avoids the one-sidedness of evaluation caused by single-dimensional features or simple models and improves the accuracy of surge protection device performance evaluation.
[0080] Please see Figure 2 , Figure 2 This is a flowchart illustrating the steps of a surge protection device nanosecond pulse test data processing method provided in Embodiment 2 of the present invention.
[0081] This invention provides a method for processing nanosecond pulse test data of surge protection devices, comprising:
[0082] Step 201: Obtain multiple training pulse parameters, perform data preprocessing on each training pulse parameter, and obtain the corresponding time series feature set.
[0083] In this embodiment of the invention, a second pulse injection experiment is conducted using a preset second pulse injection experimental platform to obtain multiple training pulse parameters (training pulse parameters include pulse parameters and original voltage waveform data, such as...). Figure 4 As shown in the figure), the original waveform data of the current is used to perform data preprocessing (i.e., three-level preprocessing and multi-scale feature extraction) on each training pulse parameter to obtain the corresponding time series feature set.
[0084] It is worth mentioning that during the second pulse injection experiment, the parameters of the surge protection device can be determined according to Table 1. The surge protection device includes MOV type surge protection device, TVS type surge protection device, and TSS type surge protection device.
[0085] Table 1
[0086]
[0087] It should be noted that, for reference Figure 5 As shown, the specific setup for the second pulse injection experiment is as follows: the DUT is a surge protector, and Rp is a 150-ohm current-limiting resistor, mainly used to protect the circuit. The dashed box shows the internal structure of the pulse source. The specific experimental procedure for the second pulse injection is as follows: the output voltage of the pulse source is adjusted to 10kV, 20kV, and 30kV respectively. Figure 5 At point V1, a resistor divider (division ratio 153:0.521) was used to measure the output voltage of the pulse source. At point V2, a NorthStar PVM-6 high-voltage probe (50kHz~500MHz, 0~100kV) was used to measure the voltage across the surge protection device. At point C1, a Pearson 8585C current probe (50kHz~500MHz, 0~500A) was used to measure the device current. The measuring equipment was connected to an MSO44 oscilloscope (1GHz bandwidth, sampling rate not less than 2GSa / s) via a measuring cable to obtain the corresponding raw voltage and current waveform data.
[0088] Step 202: Train the preset initial feature prediction model using the temporal feature set to obtain the corresponding target feature prediction model.
[0089] Further, step 202 includes the following sub-steps:
[0090] S11. Input the time series feature set into the preset initial feature prediction model for training to obtain the corresponding training feature sequence data.
[0091] Training feature sequence data refers to the predicted feature sequence obtained by inputting the time series feature set into the initial feature prediction model.
[0092] In this embodiment of the invention, a time-series feature set is used as input to a preset initial feature prediction model for training, thereby obtaining the corresponding training feature sequence data.
[0093] S12. Based on the preset loss function, calculate the loss function value of the time series feature set according to the training feature sequence data.
[0094] In this embodiment of the invention, the training feature sequence data and the time-series feature set are input into a preset loss function to obtain the corresponding loss function value.
[0095] It should be noted that the loss function is as follows:
[0096]
[0097] in, The value of the loss function. The total number of samples, The total dimension of the features. Let d be the weight of the d-th feature dimension. Let be the true label value of the b-th sample in the d-th feature dimension. This is the model prediction value for the b-th sample in the d-th feature dimension, where b is the index of the sample and d is the index of the feature dimension.
[0098] S13. When the loss function value is greater than or equal to the preset loss threshold, adjust the network parameters of the initial feature prediction model until the loss function value is less than the loss threshold.
[0099] The loss threshold refers to the maximum allowable value of the loss function that is preset according to the accuracy requirements of surge protection device test data processing, and is set to 0.05.
[0100] In this embodiment of the invention, when the loss function value is greater than or equal to 0.05, the network parameters of the initial feature prediction model are adjusted by gradient descent until the loss function value is less than 0.05.
[0101] S14. When the loss function value is less than the loss threshold, a target feature prediction model is generated.
[0102] In this embodiment of the invention, when the loss function value is less than 0.05, a target feature prediction model is generated.
[0103] Step 203: Obtain the nanosecond pulse test data of the surge protection device, and perform feature prediction on the nanosecond pulse test data through the target feature prediction model to obtain the corresponding multi-scale time series feature sequence.
[0104] Further, see Figure 3 As shown, the target feature prediction model includes a preprocessing network, a feature decoupling network, and a temporal attention fusion layer. Step 103 includes the following sub-steps:
[0105] S21. The nanosecond pulse test data is preprocessed through a preprocessing network to obtain the corresponding data feature vector. The preprocessing network includes a first fully connected layer and a second fully connected layer connected in sequence.
[0106] It should be noted that the first fully connected layer refers to a fully connected layer with an embedded ReLU activation function. The second fully connected layer refers to a fully connected layer with an embedded LeakyReLU activation function.
[0107] In this embodiment of the invention, the nanosecond pulse test data is preprocessed by a first fully connected layer and a second fully connected layer connected in sequence to obtain the corresponding data feature vector.
[0108] It is worth mentioning that the nonlinear representation of nanosecond pulse test data can be enhanced by using a preprocessing network.
[0109] S22. Multi-scale feature extraction is performed on the data feature vector through a feature decoupling network to obtain time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector.
[0110] Further, see Figure 3 As shown, S22 includes the following sub-steps:
[0111] S221. Temporal feature extraction is performed on the data feature vector through the temporal feature branch to obtain the corresponding temporal feature vector. The temporal feature branch includes a first fully connected layer and a bidirectional long short-term memory network connected in sequence.
[0112] In this embodiment of the invention, the data feature vector is extracted in the temporal domain by sequentially connecting a first fully connected layer and a bidirectional LSTM to obtain the corresponding temporal feature vector.
[0113] S222. Frequency domain feature extraction is performed on the data feature vector through the frequency domain feature branch to obtain the corresponding frequency domain feature vector. The frequency domain feature branch includes a second fully connected layer, a one-dimensional convolutional layer, a global average pooling layer and a third fully connected layer connected in sequence.
[0114] It should be noted that the kernel size of the one-dimensional convolutional layer is 3. The third fully connected layer is a fully connected layer with an embedded Linear activation function.
[0115] In this embodiment of the invention, frequency domain features are extracted from the data feature vector by sequentially connecting a second fully connected layer, a one-dimensional convolutional layer, a global average pooling layer, and a third fully connected layer, to obtain the corresponding frequency domain feature vector.
[0116] S223. Extract time-frequency domain features from the data feature vector through the time-frequency domain feature branch to obtain the corresponding time-frequency domain feature vector. The time-frequency domain feature branch includes a first fully connected layer, a wavelet packet transform layer, an attention weight layer, and a third fully connected layer connected in sequence.
[0117] In this embodiment of the invention, time-frequency domain features are extracted from the data feature vector by sequentially connecting a first fully connected layer, a wavelet packet transform layer, an attention weight layer, and a third fully connected layer, resulting in the corresponding time-frequency domain feature vector. The wavelet packet transform layer has a fixed decomposition layer number of 3 and outputs energy in 8 frequency bands.
[0118] It is worth mentioning that by connecting the wavelet packet transform layer and the attention weight layer, the wavelet packet energy output is correlated with the pulse rise time, and the energy ratio of the dominant frequency band of the rising edge is learned by strengthening the attention weight layer.
[0119] S23. The temporal attention fusion layer is used to fuse the temporal feature vector, frequency feature vector and time-frequency domain feature vector to obtain the corresponding multi-scale temporal feature sequence.
[0120] In this embodiment of the invention, a temporal attention fusion layer is used to fuse the temporal feature vector, frequency domain feature vector, and time-frequency domain feature vector. (The temporal attention fusion layer calculates the attention weights corresponding to the temporal feature vector, frequency domain feature vector, and time-frequency domain feature vector respectively. For example, the attention weight of the temporal feature vector = Sigmoid activation function (temporal feature vector), the attention weight of the frequency domain feature vector == Sigmoid activation function (frequency domain feature vector), and the attention weight of the time-frequency domain feature vector == Sigmoid activation function (time-frequency domain feature vector). The temporal feature vector, frequency domain feature vector, and time-frequency domain feature vector are then weighted and fused using each attention weight.) This yields the corresponding multi-scale temporal feature sequence.
[0121] Step 204: Using the least squares method, linearly fit the pre-acquired historical multi-scale time series feature sequence set and multi-scale time series feature sequence to obtain the corresponding trend term slope.
[0122] The historical multi-scale time series feature sequence set refers to the multi-scale time series feature sequence obtained in the first few sampling periods.
[0123] In this embodiment of the invention, based on the least squares method, a linear fit is performed on the pre-acquired historical multi-scale time series feature sequence set and the multi-scale time series feature sequence to obtain the corresponding trend term slope.
[0124] It is worth mentioning that the least squares method is a mathematical optimization method used to solve the parameters of a linear regression model. The core idea is to determine the coefficients (such as the slope of the trend term) in the model by minimizing the sum of squared errors between the actual observed values of the dependent variable and the predicted values of the model.
[0125] Step 205: Compare the multi-scale time series feature sequence with the corresponding failure precursor judgment threshold to obtain the corresponding comparison result.
[0126] The failure precursor judgment threshold refers to the standard value of each feature dimension in a multi-scale time series feature sequence.
[0127] In this embodiment of the invention, it is determined whether each feature dimension in the multi-scale time-series feature sequence does not exceed the corresponding feature standard value. If any feature dimension exceeds the corresponding feature standard value, the comparison result is determined to indicate that the device has a precursor to failure. If none of the feature dimensions exceed the corresponding feature standard value, the comparison result is determined to indicate that the device is normal.
[0128] It should be noted that the feature dimension refers to the rising edge steepness, oscillation attenuation coefficient, current and voltage synchronization, stable segment fluctuation, post-pulse recovery time, cumulative energy, feature frequency, bandwidth, high-frequency energy ratio, wavelet packet energy entropy, and rising edge dominant frequency band in a multi-scale time series feature sequence.
[0129] It is worth mentioning that the expression for the feature standard value is: Feature standard value = Mean value of each feature dimension in the normal stage + Safety margin coefficient * (Standard deviation of each feature dimension in the normal stage).
[0130] Step 206: When the comparison result indicates that the device has signs of failure, determine whether the slope of the trend term is greater than the preset slope threshold.
[0131] In this embodiment of the invention, when the comparison result indicates that the device has signs of impending failure, it is determined whether the slope of the trend term exceeds a preset slope threshold.
[0132] Step 207: When the slope of the trend term is greater than the slope threshold, the analysis result is determined to be a device anomaly.
[0133] In this embodiment of the invention, when the slope of the trend term exceeds the slope threshold, the analysis result is determined to be a device anomaly.
[0134] It should be noted that the specific process for configuring the slope threshold is as follows:
[0135] C1. Obtain the slopes of multiple standard trend terms, and average the slopes of each standard trend term to obtain the corresponding first mean.
[0136] The standard trend slope refers to the slope of the trend term within the normal phase.
[0137] In this embodiment of the invention, the slopes of all trend terms within the normal phase are obtained, and the slopes of each trend term are averaged to obtain the corresponding first mean.
[0138] C2. Multiply the preset safety margin coefficient by the standard deviation of the slope of each standard trend term to obtain the corresponding first multiplication value.
[0139] The safety margin factor refers to a margin factor set based on engineering experience for judging the precursors of failure. Its value is typically between 2 and 3.
[0140] In this embodiment of the invention, the standard deviation of the slope of all trend terms in the normal phase is calculated, and the multiplication between the preset safety margin coefficient and the standard deviation is calculated to obtain the corresponding first multiplication value.
[0141] C3. Sum the first mean and the first multiplier to obtain the corresponding slope threshold.
[0142] In this embodiment of the invention, the sum between the first mean and the first multiplier is calculated to obtain the corresponding slope threshold.
[0143] Step 208: When the slope of the trend term is less than or equal to the slope threshold, the analysis result is determined to be that the device is normal.
[0144] In this embodiment of the invention, when the slope of the trend term does not exceed the slope threshold, the analysis result is determined to be that the device is normal.
[0145] Step 209: When the comparison result indicates that the device is normal, the analysis result is confirmed as the device being normal.
[0146] In this embodiment of the invention, when the comparison result indicates that the device is normal, the comparison result is determined as the analysis result.
[0147] In this embodiment of the invention, multiple training pulse parameters are acquired, and data preprocessing is performed on each training pulse parameter to obtain a corresponding time-series feature set. The time-series feature set is then used to train a preset initial feature prediction model to obtain a corresponding target feature prediction model. Nanosecond pulse test data of the surge protection device is acquired, and feature prediction is performed on the nanosecond pulse test data using the target feature prediction model to obtain a corresponding multi-scale time-series feature sequence. Based on a pre-acquired set of historical multi-scale time-series feature sequences, failure precursor analysis is performed on the multi-scale time-series feature sequence to obtain the corresponding analysis results. This overcomes the technical problem that existing data processing methods often employ simple noise filtering or baseline correction methods, which cannot accurately reflect the true working state of the surge protection device and reduce the accuracy of surge protection device performance evaluation. This invention preprocesses the parameters of each training pulse to obtain the corresponding time-series feature set, providing a high-quality time-series feature set for the training of the subsequent initial feature prediction model. Then, the target feature prediction model is used to predict the features of the nanosecond pulse test data to obtain the corresponding multi-scale time-series feature sequence. Based on the pre-acquired historical multi-scale time-series feature sequence set, the multi-scale time-series feature sequence is analyzed for failure precursors to obtain the corresponding analysis results. This avoids the one-sidedness of evaluation caused by single-dimensional features or simple models and improves the accuracy of surge protection device performance evaluation.
[0148] Please see Figure 6 , Figure 6 This is a structural block diagram of a surge protection device nanosecond pulse test data processing system provided in Embodiment 3 of the present invention.
[0149] This invention provides a nanosecond pulse test data processing system for surge protection devices, comprising:
[0150] Preprocessing module 301 is used to acquire multiple training pulse parameters, perform data preprocessing on each training pulse parameter, and obtain the corresponding time series feature set;
[0151] Training module 302 is used to train a preset initial feature prediction model using a temporal feature set to obtain the corresponding target feature prediction model;
[0152] Prediction module 303 is used to acquire nanosecond pulse test data of surge protection devices, and to perform feature prediction on nanosecond pulse test data through target feature prediction model to obtain corresponding multi-scale time series feature sequences;
[0153] The failure precursor analysis module 304 is used to perform failure precursor analysis on multi-scale time series feature sequences based on a pre-acquired set of historical multi-scale time series feature sequences, and obtain the corresponding analysis results.
[0154] Furthermore, training module 302 includes:
[0155] The training submodule is used to input the time series feature set into the preset initial feature prediction model for training, and obtain the corresponding training feature sequence data.
[0156] The loss submodule is used to calculate the loss function value of the time-series feature set based on the training feature sequence data, according to a preset loss function.
[0157] The analysis submodule is used to adjust the network parameters of the initial feature prediction model when the loss function value is greater than or equal to the preset loss threshold, until the loss function value is less than the loss threshold.
[0158] When the loss function value is less than the loss threshold, a target feature prediction model is generated.
[0159] Furthermore, the target feature prediction model includes a preprocessing network, a feature decoupling network, and a temporal attention fusion layer. The prediction module 303 includes:
[0160] The preprocessing submodule is used to preprocess the nanosecond pulse test data through a preprocessing network to obtain the corresponding data feature vector. The preprocessing network includes a first fully connected layer and a second fully connected layer connected in sequence.
[0161] The decoupling submodule is used to perform multi-scale feature extraction on the data feature vector through the feature decoupling network to obtain time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector.
[0162] The feature fusion submodule is used to fuse time-domain feature vectors, frequency-domain feature vectors, and time-frequency-domain feature vectors through a temporal attention fusion layer to obtain the corresponding multi-scale temporal feature sequence.
[0163] Furthermore, the feature decoupling network includes a time-domain feature branch, a frequency-domain feature branch, and a time-frequency-domain feature branch. The decoupling submodule includes:
[0164] The temporal feature extraction unit is used to extract temporal features from the data feature vector through the temporal feature branch to obtain the corresponding temporal feature vector. The temporal feature branch includes a first fully connected layer and a bidirectional long short-term memory network connected in sequence.
[0165] The frequency domain feature extraction unit is used to extract frequency domain features from the data feature vector through the frequency domain feature branch to obtain the corresponding frequency domain feature vector. The frequency domain feature branch includes a second fully connected layer, a one-dimensional convolutional layer, a global average pooling layer and a third fully connected layer connected in sequence.
[0166] The time-frequency domain feature extraction unit is used to extract time-frequency domain features from the data feature vector through the time-frequency domain feature branch to obtain the corresponding time-frequency domain feature vector. The time-frequency domain feature branch includes a first fully connected layer, a wavelet packet transform layer, an attention weight layer and a third fully connected layer connected in sequence.
[0167] Furthermore, the failure precursor analysis module 304 includes:
[0168] The fitting submodule is used to perform linear fitting on the pre-acquired historical multi-scale time series feature sequence set and multi-scale time series feature sequence using the least squares method to obtain the corresponding trend term slope.
[0169] The comparison submodule is used to compare the multi-scale time series feature sequences with the corresponding failure precursor judgment thresholds to obtain the corresponding comparison results;
[0170] When the comparison result indicates that the device has signs of failure, it is determined whether the slope of the trend term is greater than the preset slope threshold.
[0171] When the slope of the trend term is greater than the slope threshold, the analysis result is determined to be a device anomaly.
[0172] When the slope of the trend term is less than or equal to the slope threshold, the analysis result is determined to be that the device is normal.
[0173] If the comparison result indicates that the device is normal, then the analysis result is determined to be that the device is normal.
[0174] Furthermore, the specific process for configuring the slope threshold is as follows:
[0175] Obtain the slopes of multiple standard trend terms, and average the slopes of each standard trend term to obtain the corresponding first mean.
[0176] The first multiplication value is obtained by multiplying the preset safety margin coefficient with the standard deviation of the slope of each standard trend term.
[0177] The first mean and the first multiplier are summed to obtain the corresponding slope threshold.
[0178] Please see Figure 7 , Figure 7 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention.
[0179] An electronic device according to an embodiment of the present invention includes: a memory 401 and a processor 402. The memory 401 stores a computer program. When the computer program is executed by the processor 402, the processor 402 executes the surge protection device nanosecond pulse test data processing method as described in any of the above embodiments.
[0180] Memory 401 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 401 has storage space 403 for program code 413 for performing any of the method steps described above. For example, storage space 403 for program code may include individual program codes 413 for implementing the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When this code is run by a computing device, it causes the computing device to perform the various steps in the surge protection device nanosecond pulse test data processing method described above.
[0181] Embodiment 5 of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the surge protection device nanosecond pulse test data processing method as described in any of the above embodiments.
[0182] Embodiment 6 of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the surge protection device nanosecond pulse test data processing method as described in any of the above embodiments.
[0183] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0184] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.
[0185] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0186] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0187] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0188] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for processing nanosecond pulse test data of surge protection devices, characterized in that, include: Multiple training pulse parameters are acquired, and data preprocessing is performed on each training pulse parameter to obtain the corresponding time series feature set; The preset initial feature prediction model is trained using the time-series feature set to obtain the corresponding target feature prediction model; The nanosecond pulse test data of the surge protection device is acquired, and the target feature prediction model is used to predict the features of the nanosecond pulse test data to obtain the corresponding multi-scale time series feature sequence. Based on the pre-acquired historical multi-scale time series feature sequence set, failure precursor analysis is performed on the multi-scale time series feature sequence to obtain the corresponding analysis results.
2. The method for processing nanosecond pulse test data of surge protection devices according to claim 1, characterized in that, The step of training a preset initial feature prediction model using the temporal feature set to obtain a corresponding target feature prediction model includes: The time-series feature set is input into a preset initial feature prediction model for training to obtain the corresponding training feature sequence data; Based on a preset loss function, the loss function value of the time-series feature set is calculated according to the training feature sequence data; When the loss function value is greater than or equal to a preset loss threshold, the network parameters of the initial feature prediction model are adjusted until the loss function value is less than the loss threshold. When the loss function value is less than the loss threshold, a target feature prediction model is generated.
3. The method for processing nanosecond pulse test data of surge protection devices according to claim 1, characterized in that, The target feature prediction model includes a preprocessing network, a feature decoupling network, and a temporal attention fusion layer. The step of using the target feature prediction model to predict features from the nanosecond pulse test data to obtain the corresponding multi-scale temporal feature sequence includes: The nanosecond pulse test data is preprocessed by the preprocessing network to obtain the corresponding data feature vector. The preprocessing network includes a first fully connected layer and a second fully connected layer connected in sequence. The feature decoupling network is used to perform multi-scale feature extraction on the data feature vector to obtain time-domain feature vector, frequency-domain feature vector and time-frequency-domain feature vector; The temporal attention fusion layer fuses the temporal feature vector, the frequency feature vector, and the time-frequency domain feature vector to obtain the corresponding multi-scale temporal feature sequence.
4. The method for processing nanosecond pulse test data of surge protection devices according to claim 3, characterized in that, The feature decoupling network includes a time-domain feature branch, a frequency-domain feature branch, and a time-frequency-domain feature branch. The step of performing multi-scale feature extraction on the data feature vector through the feature decoupling network to obtain the time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector includes: Temporal features are extracted from the data feature vector through the temporal feature branch to obtain the corresponding temporal feature vector. The temporal feature branch includes a first fully connected layer and a bidirectional long short-term memory network connected in sequence. The frequency domain feature vector is obtained by extracting frequency domain features from the data feature vector through the frequency domain feature branch. The frequency domain feature branch includes a second fully connected layer, a one-dimensional convolutional layer, a global average pooling layer and a third fully connected layer connected in sequence. The time-frequency domain feature vector is extracted by the time-frequency domain feature branch to obtain the corresponding time-frequency domain feature vector. The time-frequency domain feature branch includes a first fully connected layer, a wavelet packet transform layer, an attention weight layer and a third fully connected layer connected in sequence.
5. The method for processing nanosecond pulse test data of surge protection devices according to claim 1, characterized in that, The step of performing failure precursor analysis on the multi-scale time-series feature sequences based on a pre-acquired historical multi-scale time-series feature sequence set to obtain the corresponding analysis results includes: The least squares method is used to perform linear fitting on the pre-acquired historical multi-scale time series feature sequence set and the multi-scale time series feature sequence to obtain the corresponding trend term slope. The multi-scale time-series feature sequence is compared with the corresponding failure precursor judgment threshold to obtain the corresponding comparison result; When the comparison result indicates that the device has signs of impending failure, it is determined whether the slope of the trend term is greater than a preset slope threshold. When the slope of the trend term is greater than the slope threshold, the analysis result is determined to be a device anomaly. When the slope of the trend term is less than or equal to the slope threshold, the analysis result is determined to be that the device is normal. When the comparison result indicates that the device is normal, the analysis result is determined to be that the device is normal.
6. The method for processing nanosecond pulse test data of surge protection devices according to claim 5, characterized in that, The specific process for configuring the slope threshold is as follows: Obtain the slopes of multiple standard trend terms, and average the slopes of each standard trend term to obtain the corresponding first mean. The preset safety margin coefficient is multiplied by the standard deviation of the slope of each of the standard trend terms to obtain the corresponding first multiplication value. The first mean and the first multiplier are summed to obtain the corresponding slope threshold.
7. A surge protection device nanosecond pulse test data processing system, characterized in that, include: The preprocessing module is used to acquire multiple training pulse parameters, perform data preprocessing on each training pulse parameter, and obtain the corresponding time series feature set; The training module is used to train the preset initial feature prediction model using the time-series feature set to obtain the corresponding target feature prediction model. The prediction module is used to acquire nanosecond pulse test data of surge protection devices, and to perform feature prediction on the nanosecond pulse test data through the target feature prediction model to obtain the corresponding multi-scale time series feature sequence. The failure precursor analysis module is used to perform failure precursor analysis on the multi-scale time series feature sequences based on a pre-acquired set of historical multi-scale time series feature sequences, and obtain the corresponding analysis results.
8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the surge protection device nanosecond pulse test data processing method as described in any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the surge protection device nanosecond pulse test data processing method as described in any one of claims 1-6.
10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the surge protection device nanosecond pulse test data processing method as described in any one of claims 1-6.