Distribution network type lightning arrester leakage current optimization measurement method

Abstract

The invention provides a distribution network type lightning arrester leakage current optimization measurement method. The method comprises the steps of signal acquisition, preliminary signal denoising and preprocessing, wavelet transform signal decomposition, high-frequency noise component elimination and processing, input convolutional neural network CNN descrambling, integrated learning model optimization, output of descrambled leakage current signals, performance evaluation and result feedback. Through a multi-level processing method combining wavelet transform and a convolutional neural network, high-frequency noise in a weak leakage current signal is effectively removed, effective characteristics of the signal are reserved in the denoising process, and compared with a traditional method depending on a single signal processing technology, signal descrambling can be more accurate and effective, and the signal descrambling efficiency is improved. According to the wavelet transform signal decomposition method, the signal can be decomposed into components with different frequencies, high-frequency noise components are effectively eliminated, low-frequency and intermediate-frequency effective signals are reserved, then the convolutional neural network further improves the definition and reliability of the signal through depth feature extraction, and accurate detection of the leakage current signal is ensured.

Description

Technical Field

[0001] The present invention relates to the field of distribution networks, and particularly to an optimized measurement method for the leakage current of distribution network type lightning arresters. Background Art

[0002] With the continuous development of smart grid technology, distribution network type lightning arresters play an increasingly important role in the power system, especially in protecting grid equipment from lightning overvoltage. However, with the increasing complexity of the distribution network environment, the accuracy of leakage current measurement has become an urgent problem to be solved. In particular, factors such as electromagnetic interference and noise pollution often make it difficult for traditional leakage current measurement methods to be effectively applied in complex power environments.

[0003] Currently, the measurement of the leakage current of distribution network type lightning arresters mostly relies on traditional current sensors. When facing multiple interferences in the grid environment, the accuracy of the measurement results often fails to meet the requirements. Especially in high-noise and high-interference situations, the measurement error is relatively large. In addition, although existing signal processing methods can alleviate the influence of some noise, for complex electromagnetic environments and dynamically changing signals, traditional denoising techniques still seem inadequate, resulting in limited improvement in measurement accuracy.

[0004] Therefore, in recent years, the research on optimizing the measurement of leakage current by combining wavelet transform and deep learning technology has gradually become a hot topic. Wavelet transform can effectively separate the effective information and noise components in the signal, while the multi-layer convolutional neural network CNN performs excellently in the field of signal processing with its powerful automatic feature extraction ability. However, the combined and optimized application of such technologies is still in the exploratory stage, lacking a systematic comprehensive solution. Especially in the actual application of measuring the leakage current of distribution network type lightning arresters, how to improve the accuracy and robustness of signal processing remains an urgent problem to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide an optimized measurement method for the leakage current of distribution network type lightning arresters in view of the above-mentioned deficiencies of the prior art, providing an efficient, accurate and reliable signal denoising and leakage current measurement solution, which can be widely applied to the leakage current monitoring of lightning arresters in distribution network systems.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: The present invention provides an optimized measurement method for the leakage current of distribution network type lightning arresters, including the following steps: The purpose of the present invention is to provide an optimized measurement method for the leakage current of distribution network type lightning arresters in view of the above-mentioned deficiencies of the prior art, providing an efficient, accurate and reliable signal denoising and leakage current measurement solution, which can be widely applied to the leakage current monitoring of lightning arresters in distribution network systems.

[0007] To achieve the above object, the present invention adopts the following technical solutions: The present invention provides a method for optimizing the measurement of leakage current of a distribution network type lightning arrester, including the following steps: S1. Signal acquisition: Construct a device for generating a weak leakage current signal under a strong interference background, and the sending device is used to simulate the actual leakage current signal in the power grid and collect data; S2. Preliminary signal denoising and preprocessing: Laying a foundation for wavelet transform and convolutional neural network deinterference; S3. Wavelet transform signal decomposition: Perform wavelet transform on the weak current signal, extract different frequency components of the signal, and provide an effective signal basis for subsequent denoising and feature extraction; S4. Elimination and processing of high-frequency noise components; S5. Input the convolutional neural network CNN for deinterference: The signal after wavelet transform and noise processing is used as input data and transmitted to the convolutional neural network; S6. Optimization of the ensemble learning model; S7. Output the deinterfered leakage current signal: After deinterference by the convolutional neural network and optimization by the ensemble learning model, the obtained signal is further processed, and finally the deinterfered leakage current signal is output; S8. Performance evaluation and result feedback.

[0008] Further, the specific content of S2 is as follows: Using a band-pass filter can remove low-frequency and high-frequency noises and retain the effective components of the target signal. The transfer function of the band-pass filter is expressed as: ; Among them, is the complex frequency variable; is the center frequency, indicating the center of the working frequency band of the filter; is the quality factor, which determines the bandwidth of the filter; When performing pseudo-signal elimination, a statistical-based method is used to detect outliers, and the specific formula is: ; Among them, is the value of the current signal; is the mean value of the signal; is the standard deviation of the signal.

[0009] Further, the specific content of S3 is as follows: S301. Adopt a wavelet basis function, and set the number of layers of wavelet decomposition to three layers. The result formula of wavelet transform is: ; Among them, is the input leakage current signal; is the scale factor; is the translation factor; is the wavelet basis function, which is the signal for decomposition and analysis; is the time point of the signal; S302. Perform multi - layer wavelet decomposition on the input leakage current signal to obtain the components as follows: The low - frequency component and high - frequency component of the first layer; The low - frequency component and high - frequency component of the second layer; The low - frequency component and high - frequency component of the third layer; Among them, is the one - layer decomposition component of the leakage current signal wavelet transform; is the two - layer decomposition component of the leakage current signal wavelet transform; is the three - layer decomposition component of the leakage current signal wavelet transform; , and are all low - frequency components; , and are all high - frequency components.

[0010] Furthermore, the specific content of S4 is as follows: Use the soft - threshold algorithm to remove the high - frequency noise components: ; Among them, is the soft - threshold function; is the set threshold; is the wavelet coefficient; Reconstruct the input leakage current signal to obtain the denoised signal: ; is the denoised leakage current signal; is the wavelet coefficient after denoising processing; is the index of the wavelet coefficient.

[0011] Furthermore, the specific content of S5 is as follows: The convolution operation scans the input signal through multiple convolution kernels to generate a feature map. Each convolution kernel can learn a specific pattern in the signal. The formula is: ; Among them, is the output feature map of the convolution layer; is the th convolution kernel; is the bias term of the convolutional kernel; is the number of convolutional kernels; The pooling layer is used to reduce the spatial dimension of the feature map, thereby reducing the computational amount, and can retain the information in the feature map. The dimensionality reduction is performed by taking the maximum value in the local area. The formula for max pooling is: ; where, is the output signal after pooling; is the pooling window; is the signal value within the window; The signal features passing through the convolutional layer and the pooling layer will be unfolded and passed to the fully connected layer. The fully connected layer maps the features to the output space. The calculation formula for the final signal de-noising fully connected layer is: ; where, is the activation function; is the weight matrix; is the signal feature after convolution and pooling; is the bias term; In order to measure the de-noising effect, the mean square error is used as the loss function: ; where, is the number of signal samples; is the true signal value; is the predicted signal value; is the index of the sample, from 1 to represents each individual sample in the dataset; The training of the convolutional neural network CNN is achieved through multiple iterations. Each iteration updates the weights of the network according to the error. The training process is carried out by batch gradient descent, and an optimization algorithm is used to accelerate convergence. The update rule of the optimization algorithm is: ; where, and are the estimates of the first and second moments respectively; and are the momentum terms; is the learning rate; is the gradient operator; is a small constant to prevent division by zero error; is the update parameter; is the bias correction term of the first moment; is the bias correction term of the second moment.

[0012] Furthermore, the specific S8 is: S801. SNR (Signal-to-Noise Ratio) evaluation. The formula for calculating the signal-to-noise ratio is as follows: ; Wherein, is the effective part of the signal; is the noise component, calculating the energy ratio of the signal and the noise; S802. During the processing, certain errors will occur. That is, the error analysis includes quantifying the difference between the de-noised signal and the actual leakage current signal. The formula for the error analysis is as follows: ; Wherein, is the error analysis function; is the hyperparameter for adjusting the high-order error; is the kurtosis for further measuring the error; S803. Delay analysis: The processing delay refers to the time interval from when the input signal enters the system to when the output signal is generated. During the processing of the leakage current signal, the smaller the delay, the higher the real-time performance of the system. The formula for calculating the processing delay is as follows: ; Wherein, is the average value of the processing delay; is the output time of the th signal sample; is the input time of the th signal sample.

[0013] Furthermore, in the S1, the generating device includes: a hardware device unit and a data processing and control unit: The hardware device unit includes: Signal generator module: responsible for generating weak leakage current signals, with a frequency range covering 0.1 HZ to 1 KHZ; Power amplifier module: enhancing weak signals, adopting a high linearity design to ensure that the input noise is less than 10 uV to guarantee low distortion during signal amplification; Multi-magnetic current TMR sensor module: responsible for accurately collecting weak leakage current signals; Oscilloscope module: used to display and read the collected signal waveforms in real time, monitoring the changes of weak leakage current signals; DC power supply module: providing stable power supply; The data processing and control unit includes: Computer and control system: communicating with the hardware device through an interface, controlling the input, output, and amplification of signals; Interference Removal Algorithm and Software Processing Module: Using wavelet transform and convolutional neural network to perform multi-level denoising and interference removal on signals; Data Storage and Analysis Module: Responsible for storing the original collected data, the interference-removed signal data, and various performance evaluation results to ensure the long-term preservation and backup of the data; Output Module: Real-time display the interference-removed leakage current signal on the screen, providing intuitive waveforms and curves to help operators understand the system status.

[0014] The beneficial effects of the present invention are as follows: Through the multi-level processing method combining wavelet transform and convolutional neural network, the high-frequency noise in the weak leakage current signal is effectively removed, and the effective features of the signal are retained during the denoising process. Compared with the traditional method relying on a single signal processing technology, the signal interference removal can be more accurate and effective. The wavelet transform signal decomposition method can decompose the signal into components of different frequencies, effectively removing the high-frequency noise components and retaining the effective signals of low and medium frequencies. Subsequently, the convolutional neural network further improves the clarity and reliability of the signal through deep feature extraction, ensuring the accurate detection of the leakage current signal.

[0015] In addition, the interference-removed signal is optimized through the integrated learning random forest model, further improving the accuracy of the signal. This multi-level signal optimization processing can not only effectively remove interference but also improve the measurement accuracy of the leakage current signal, avoiding misjudgment caused by noise interference; it can perform high-precision leakage current measurement in a complex interference environment, and the robustness and stability of the system are significantly improved. Especially in practical applications, it can operate stably for a long time, adapt to signal changes in different power grid environments, and ensure the reliability of the system. Brief Description of the Drawings

[0016] Figure 1 It is a flowchart of an optimized measurement method for the leakage current of a distribution network type lightning arrester. Detailed Embodiments

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Please refer to Figure 1 , an optimized measurement method for the leakage current of a distribution network type lightning arrester, comprising the following steps: S1. Signal Acquisition; Construct a device for generating a weak leakage current signal under a strong interference background, and the sending device is used to simulate the actual leakage current signal in the power grid and collect data; S2. Preliminary Signal Denoising and Preprocessing: Laying the foundation for interference removal by wavelet transform and convolutional neural network; S3. Wavelet transform signal decomposition: Perform wavelet transform on the weak current signal to extract different frequency components of the signal, providing an effective signal basis for subsequent denoising and feature extraction; S4. Rejection and processing of high-frequency noise components; S5. Input the denoised signal into the convolutional neural network CNN: The signal after wavelet transform and noise processing is used as input data and transmitted to the convolutional neural network; S6. Optimization of the ensemble learning model; Specifically, first use the random forest as the basic learning model to perform multiple trainings and predictions on the denoised signal respectively; The random forest is integrated by constructing multiple decision trees. The generation of each tree depends on a randomly selected subset of feature subsets and sample subsets, and finally the output result is determined through a voting mechanism. During the training process, the ensemble model will be optimized according to the prediction errors of each sample in the training data. By adopting the sequential ensemble learning method Boosting or the parallel ensemble learning method Bagging strategy, the weights of each basic learner are continuously adjusted, and the bias and variance of the model are gradually reduced through the error feedback mechanism.

[0019] S7. Output the denoised leakage current signal: After being denoised by the convolutional neural network and optimized by the ensemble learning model, the obtained signal is further processed and finally the denoised leakage current signal is output; The output denoised leakage current signal will be transmitted to the display terminal and the alarm system, providing accurate data support for the real-time monitoring and alarm of the system; Specifically: The denoised leakage current signal will be transmitted to the display terminal through the data interface. The display terminal is a visual monitoring system that displays the processed leakage current data in real time. Such terminals can use touch screens, LCD displays, and have functions of real-time data update and curve display. Monitoring personnel can intuitively understand the working status of lightning arresters in the distribution network based on the leakage current signal displayed on the terminal and judge whether there is abnormal leakage; The denoised leakage current signal also needs to be transmitted to the alarm system. The alarm system makes a judgment based on a preset threshold. Once the leakage current exceeds the normal range, the alarm system will immediately issue an alarm. The alarm methods usually include sound alarms, flash alarms, and remote notifications, etc. Through the alarm system, staff can take timely measures to avoid potential safety hazards such as equipment damage or electrical fires caused by abnormal leakage current; The denoised leakage current signal also needs to be stored in the database for querying, analyzing, and generating reports of historical data. The data storage system should have efficient data management capabilities, be able to support regular data backup and recovery functions. Through long-term data accumulation, the system can also discover potential leakage trends through data analysis and arrange early warnings and maintenance plans in advance.

[0020] S8, Performance Evaluation and Result Feedback.

[0021] In one example, the core task of preliminary signal denoising and preprocessing is to process through a series of algorithms to remove noise, pseudo-signals, and environmental interference; Specifically, for S2: A band-pass filter can be used to remove low-frequency and high-frequency noise and retain the effective components of the target signal. The transfer function of the band-pass filter is expressed as: ; Where is the complex frequency variable; is the center frequency, representing the center of the working frequency band of the filter; is the quality factor, which determines the bandwidth of the filter; When removing pseudo-signals, a statistics-based method is used to detect outliers. The specific formula is: ; Where is the value of the current signal; is the mean of the signal; is the standard deviation of the signal.

[0022] Specifically, if exceeds a preset threshold, usually taken as 3, then the signal is considered abnormal, that is, a pseudo-signal, and needs to be removed or replaced.

[0023] Where measures whether a signal or data point significantly deviates from the overall average level. If the value is large, it may indicate that the point is abnormal or an outlier.

[0024] In the last step of signal preprocessing, the amplitude of the signal is normalized to a specific range through standardization. After completing all denoising and preprocessing steps, the signal is input into the subsequent processing module.

[0025] Specifically, for S3: S301. Adopt wavelet basis functions. The number of layers of wavelet decomposition is set to three. The result formula of wavelet transform is: ; Where is the input leakage current signal; is the scale factor; is the translation factor; is the wavelet basis function, which is the signal for decomposition and analysis; is the time point of the signal; Specifically, the wavelet basis function can be the Daubechies wavelet, which has high regularity and compact support, and can better adapt to the extraction and analysis of mutation points in current signals. The number of layers of wavelet decomposition is dynamically adjusted according to the frequency range of the input signal, usually set to three layers to ensure the separation of low-frequency signals and high-frequency noise components.

[0026] S302. Perform multi-layer wavelet decomposition on the input leakage current signal to obtain the following components:

[0027] The low-frequency component and high-frequency component of the first layer; The low-frequency component and high-frequency component of the second layer; The low-frequency component and high-frequency component of the third layer; Among them, is the component of the first-layer wavelet transform of the leakage current signal; is the component of the second-layer wavelet transform of the leakage current signal; is the component of the third-layer wavelet transform of the leakage current signal; , and are all low-frequency components; , and are all high-frequency components.

[0028] Specifically, for the multi-layer wavelet decomposition of the signal: First, the signal is decomposed into low-frequency and high-frequency components. Each layer of decomposition generates a low-frequency component and a high-frequency component, and a discrete wavelet transform is performed on the signal once.

[0029] The specific content of S4 is as follows: Use the soft threshold algorithm to remove the high-frequency noise components: ; Among them, is the soft threshold function; is the set threshold; is the wavelet coefficient; Reconstruct the input leakage current signal to obtain the denoised signal: ; is the denoised leakage current signal; is the wavelet coefficient after denoising; is the index of the wavelet coefficient.

[0030] Specifically, each layer of wavelet decomposition will obtain signal components in different frequency bands. According to the specific signal characteristics, such as the frequency range of the weak leakage current signal, it is possible to choose to retain the low-frequency part and remove the high-frequency part. The high-frequency part usually contains noise components, so an appropriate threshold can be selected to process and remove unnecessary high-frequency noise.

[0031] During the wavelet transform process, after denoising the high-frequency noise components, a suitable signal reconstruction strategy is selected according to the energy distribution of each frequency band. For the weak leakage current signal, the main trend of the signal is captured by retaining the low-frequency part, while the high-frequency part is removed to reduce noise. After wavelet transform decomposition and denoising, the reconstructed signal usually has a relatively smooth trend, and most of the high-frequency noise is removed, providing optimized input data for subsequent de-noising of the convolutional neural network.

[0032] Specifically, S5 is as follows: The convolution operation scans the input signal through multiple convolution kernels to generate a feature map. Each convolution kernel can learn specific patterns in the signal. The formula is: ; Among them, is the output feature map of the convolutional layer; is the th convolution kernel; is the bias term of the convolution kernel; is the number of convolution kernels; The pooling layer is used to reduce the spatial dimension of the feature map, thereby reducing the computational amount, and can retain the information in the feature map. The dimensionality reduction is performed by taking the maximum value in the local area. The formula for max pooling is: ; Among them, is the output signal after pooling; is the pooling window; is the signal value within the window; The signal features after the convolutional layer and the pooling layer will be unfolded and passed to the fully connected layer. The fully connected layer maps the features to the output space. The calculation formula for the final signal de-noising fully connected layer is: ; Among them, is the activation function; is the weight matrix; is the signal feature after convolution and pooling; is the bias term; In order to measure the de-noising effect, the mean square error is used as the loss function: ; Among them, is the number of signal samples; is the true signal value; is the predicted signal value; is the index of the sample, ranging from 1 to indicating each individual sample in the dataset; The training of the convolutional neural network CNN is achieved through multiple iterations. In each iteration, the weights of the network are updated according to the error. The training process is carried out through batch gradient descent and an optimization algorithm is used to accelerate convergence. The update rule of the optimization algorithm is: ; Among them, and are the estimates of the first and second moments respectively; and are the momentum terms; is the learning rate; is the gradient operator; is a small constant to prevent division by zero errors; is the update parameter; is the bias correction term for the first moment; is the bias correction term for the second moment.

[0033] Specifically, the convolutional neural network CNN performs deep feature extraction on the signal through multiple layers of convolutional operations. In this process, the convolutional layer in the network first performs local perception on the input signal and extracts features from the signal through a sliding convolutional kernel. Each convolutional layer generates a feature map that can capture different frequency and time-domain features in the signal. To enhance the expressive power of the model, the network introduces non-linearity through an activation function, enabling the model to learn complex signal patterns and denoising capabilities.

[0034] After the convolutional layer, a pooling layer is introduced to reduce the dimension of the signal and retain the most important feature information. The pooling operation usually adopts max pooling or average pooling to aggregate the signal values within each small region, thereby reducing the computational complexity and making the model more robust, reducing the risk of overfitting. The pooled feature map will be further integrated through a fully connected layer to fuse the locally extracted features into global information, and finally generate a denoised output signal.

[0035] Throughout the process, the goal of the convolutional neural network CNN is to automatically learn the effective features in the signal and remove the noise components. To ensure the denoising effect of the model, a loss function is used during the training process to measure the difference between the predicted signal and the true signal, and the optimization algorithm is used to continuously adjust the network weights, thereby continuously optimizing the model performance and achieving precise denoising of the signal.

[0036] Specifically, the convolutional neural network CNN consists of three convolutional layers, two pooling layers, and one fully connected layer.

[0037] The first convolutional layer: The convolutional kernel size is 3×3, and the number is 32; The second pooling layer: Max pooling is adopted, and the window size is 2×2; The third convolutional layer: The convolutional kernel size is 3×3, and the number is 64; The fully connected layer: The output dimension is 1, which is used for the final signal reconstruction.

[0038] The specific content of S8 is as follows: S801, Signal-to-noise ratio SNR evaluation. The calculation formula for the signal-to-noise ratio is: ; Among them, is the effective part of the signal; is the noise component, and the energy ratio of the signal and the noise is calculated; S802, During the processing, there will be certain errors. That is, the error analysis includes quantifying the difference between the de-noised signal and the actual leakage current signal. The formula for the error analysis is: ; Among them, is the error analysis function; is the hyperparameter for adjusting the high-order error; is the kurtosis for further measuring the error; S803, Delay analysis: The processing delay refers to the time interval between the input signal entering the system and the output signal being generated. During the leakage current signal processing, the smaller the delay, the higher the real-time performance of the system. The formula for calculating the processing delay is: ; Among them, is the average value of the processing delay; is the output time of the th signal sample; is the input time of the

[0039] In S1, the generating device includes: a hardware device unit and a data processing and control unit: The hardware device unit includes: The signal generator module: Responsible for generating weak leakage current signals, with a frequency range covering 0.1HZ to 1KHZ; The power amplifier module: Enhances weak signals, adopts a high linearity design, ensures that the input noise is less than 10 uV, to ensure low distortion during signal amplification; Multi-magnetic current TMR sensor module: responsible for accurately collecting weak leakage current signals; Oscilloscope module: used to display and read the acquired signal waveforms in real time, and monitor the changes in weak leakage current signals; DC power supply module: provides a stable power supply; The data processing and control unit includes: Computer and control system: communicates with hardware devices through interfaces, and controls the input, output, and amplification of signals; Interference removal algorithm and software processing module: uses wavelet transform and convolutional neural network to perform multi-level denoising and interference removal on signals; Data storage and analysis module: responsible for storing the original acquired data, denoised signal data, and various performance evaluation results, ensuring the long-term preservation and backup of data; Output module: displays the denoised leakage current signal on the screen in real time, provides intuitive waveforms and curve graphs to help operators understand the system status.

[0040] Specifically, first use the DC power supply module to supply power to the power amplifier, then output a sine signal from the signal generator module. This signal passes through the load at one end of the output port of the power amplifier module and returns to the other end of the output port of the power amplifier module through the multi-magnetic current TMR sensor module, forming a complete loop; finally, the oscilloscope module reads the weak leakage current signal data processed by the multi-magnetic sensor.

[0041] Experimental design and data description Signal acquisition and noise simulation: Original signal: Generate a sine wave with a fundamental frequency of 50 Hz to simulate a sine leakage current signal with an amplitude of 10 uA.

[0042] Noise-added signal: Add the following noises: High-frequency noise: 1 kHz high-frequency white noise with an amplitude of ±2 uA; Low-frequency interference: 150 Hz low-frequency noise with an amplitude of ±1 uA; Random pulse noise: Appears randomly with an amplitude of ±5 uA; The signal-to-noise ratio (SNR) of the noise-added signal is calculated as: SNR original = 15.2 dB; Initial denoising and preprocessing: Band-pass filter parameters: Center frequency fc = 50 Hz; Bandwidth Δf = 20 Hz; Quality factor Q = 5 Wavelet transform signal decomposition: Wavelet basis function: Daubechies; Decomposition layers: 3 layers; Decomposition result: Approximation coefficient (low frequency): Retain the 50 Hz fundamental wave component; Detail coefficient (high frequency): Include 1 kHz noise and interference in the pulse.

[0043] Soft threshold denoising: Threshold λ = 1.5 μA; The high-frequency noise energy is reduced by 90%, and the signal SNR is further improved.

[0044] 1.4 Convolutional Neural Network (CNN) denoising; CNN structure: Input layer: 1D signal (length 1000 sampling points); Convolutional layer: 3 layers, convolutional kernel size 3×3, number of channels 32 → 64; Pooling layer: Max pooling, window size 1×2; Fully connected layer: 1 neuron in the output layer.

[0045] Training data: Training set: 5000 groups of noisy signals and clean signals; Validation set: 1000 groups of independent data.

[0046] Training results: The loss function (MSE) drops from the initial value of 0.25 to 0.02; The SNR of the validation set reaches: 32.1 dB Optimization of the ensemble learning model: Model configuration: Base models: Random Forest (100 trees), Gradient Boosting Machine (GBM); Ensemble strategy: Weighted average (weights optimized through cross-validation).

[0047] Performance comparison is shown in Table 1 and Table 2: Table 1 shows the performance comparison of MSE and SNR for different models

[0048] Summary: The single CNN + ensemble model further reduces the MSE by 25% and increases the SNR by 2.4 dB Overall performance evaluation.

[0049] Table 2 shows the performance comparison table of SNR and processing operations at different processing stages

[0050]

[0051] By combining wavelet transform and CNN, the SNR is increased from 15.2 dB to 34.5 dB, and the noise energy is reduced by 98%.

[0052] The embodiments described above only represent the implementation modes of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims

1. An optimized measurement method for the leakage current of a distribution network type lightning arrester, characterized in that, It includes the following steps: S1. Signal acquisition: Construct a generating device for weak leakage current signals under a strong interference background. The sending device is used to simulate the actual leakage current signals in the power grid and collect data; S2. Preliminary signal denoising and preprocessing: Laying a foundation for wavelet transform and convolutional neural network deinterference; S3. Wavelet transform signal decomposition: Perform wavelet transform on the weak current signal, extract different frequency components of the signal, and provide an effective signal basis for subsequent denoising and feature extraction; S4. High-frequency noise component removal and processing; S5. Input the convolutional neural network CNN for deinterference: The signal model after wavelet transform and noise processing is used as input data and transmitted to the convolutional neural network; S6. Ensemble learning model optimization; S7. Output the deinterfered leakage current signal: After deinterference by the convolutional neural network and optimization by the ensemble learning model, the obtained signal is further processed, and finally the deinterfered leakage current signal is output; S8. Performance evaluation and result feedback.

2. The optimized measurement method for leakage current of a distribution network type lightning arrester according to claim 1, wherein The specific content of S2 is: Using a band-pass filter can remove low-frequency and high-frequency noises and retain the effective components of the target signal. The transfer function of the band-pass filter is expressed as: ； Among them, is a complex frequency variable; is the center frequency, representing the center of the operating frequency band of the filter; is the quality factor, which determines the bandwidth of the filter; When performing pseudo-signal removal, a statistical-based method is used to detect outliers. The specific formula is: ； Among them, is the value of the current signal; is the mean value of the signal; is the standard deviation of the signal.

3. A method for optimizing the measurement of the leakage current of a distribution network type lightning arrester according to claim 2, characterized in that, The specific content of S3 is: S301. Adopt wavelet basis functions, set the number of layers of wavelet decomposition to three layers, and the result formula of wavelet transform is: ； Among them, is the leakage current signal of the input; is the scale factor; is the translation factor; is the wavelet basis function, which is the signal for decomposition and analysis; is the time point of the signal; S302. Perform multi-level wavelet decomposition on the input leakage current signal The obtained components are as follows: The low-frequency component and high-frequency component of the first layer; The low-frequency component and high-frequency component of the second layer; The low-frequency component and high-frequency component of the third layer; Among them, is the first-level decomposition component of the wavelet transform of the leakage current signal; is the second-level decomposition component of the wavelet transform of the leakage current signal; is the third-level decomposition component of the wavelet transform of the leakage current signal; , and are all low-frequency components; , and are all high-frequency components.

4. A method for optimizing the measurement of the leakage current of a distribution network type lightning arrester according to claim 3, characterized in that, The specific content of S4 is: Adopt the soft threshold algorithm to remove high-frequency noise components: ； Among them, is the soft threshold function; is the set threshold; is the wavelet coefficient; Reconstruct the input leakage current signal to obtain the denoised signal: ； is the denoised leakage current signal; is the wavelet coefficient after denoising processing; is the index of the wavelet coefficient.

5. A method for optimizing the measurement of the leakage current of a distribution network type lightning arrester according to claim 4, characterized in that, The specific content of S5 is: The convolution operation scans the input signal through multiple convolution kernels to generate a feature map. Each convolution kernel can learn specific patterns in the signal. The formula is: ； Among them, is the output feature map of the convolutional layer; is the th convolutional kernel; is the bias term of the convolutional kernel; is the number of convolutional kernels; The pooling layer is used to reduce the spatial dimension of the feature map, thereby reducing the calculation amount and being able to retain the information in the feature map. The dimensionality reduction is performed by taking the maximum value in the local area. The formula for max pooling is: ； Among them, is the output signal after pooling; is the pooling window; is the signal value within the window; The signal features after the convolution layer and the pooling layer will be unfolded and transmitted to the fully connected layer. The fully connected layer maps the features to the output space. The calculation formula for the final signal deinterference fully connected layer is: ； Among them, is the activation function; is the weight matrix; is the signal feature after convolution and pooling; is the bias term; To measure the de - scrambling effect, the mean square error is used as the loss function: ； Among them, is the number of signal samples; is the true signal value; is the predicted signal value; is the index of the sample, from 1 to represents each individual sample in the dataset; The training of the convolutional neural network CNN is achieved through multiple iterations. Each iteration updates the weights of the network according to the error. The training process is carried out by batch gradient descent, and an optimization algorithm is used to accelerate convergence. The update rule of the optimization algorithm is: ； wherein, and are the estimates of the first and second moments, respectively; and are momentum terms; is the learning rate; is the gradient operator; is a small constant to prevent division-by-zero errors; is the update parameter; is the bias correction term for the first moment; is the bias correction term for the second moment.

6. The optimized measurement method for leakage current of a distribution network type lightning arrester according to claim 5, characterized in that, The specific content of S8 is: S801. Signal-to-noise ratio SNR evaluation. The calculation formula for the signal-to-noise ratio is: ； Among them, is the valid part of the signal; is the noise component, and calculate the energy ratio of the signal and the noise; S802. During the processing, a certain error will occur. That is, the error analysis includes quantifying the difference between the deinterfered signal and the actual leakage current signal. The formula for the error analysis is: ； Among them, is the error analysis function; is the hyperparameter for adjusting the high-order error; is the kurtosis for further measuring the error; S803. Delay analysis: The processing delay refers to the time interval between the input signal entering the system and the generation of the output signal. In the process of processing leakage current signals, the smaller the delay, the higher the real-time performance of the system. The formula for calculating the processing delay is: ； wherein, is the average value of processing delay; is the output time of the th signal sample; is the input time of the th signal sample.

7. A method for optimizing the measurement of the leakage current of a distribution network type lightning arrester according to claim 6, characterized in that, In S1, the generating device includes: a hardware device unit and a data processing and control unit: The hardware device unit includes: A signal generator module: Responsible for generating weak leakage current signals, with a frequency range covering 0.1HZ to 1KHZ; Power amplifier module: Enhance weak signals, adopt a high linearity design, ensure that the input noise is less than 10 uV to guarantee low distortion during signal amplification; Multi-magnetic current TMR sensor module: Responsible for accurately collecting weak leakage current signals; Oscilloscope module: Used to display and read the collected signal waveforms in real time, and monitor the changes in weak leakage current signals; DC power supply module: Provide stable power supply; The data processing and control unit includes: Computer and control system: Communicate with hardware devices through interfaces to control the input, output, and amplification of signals; Interference removal algorithm and software processing module: Use wavelet transform and convolutional neural network to perform multi-level denoising and interference removal on signals; Data storage and analysis module: Responsible for storing the original collected data, the denoised signal data, and various performance evaluation results to ensure the long-term preservation and backup of data; Output module: Display the denoised leakage current signal on the screen in real time, provide intuitive waveforms and curve graphs to help operators understand the system status.

Images