A new energy station cable terminal partial discharge signal identification method and system

Abstract

This invention discloses a method and system for identifying partial discharge signals in cable terminals of new energy power plants, comprising the following steps: acquiring the original signal and preprocessing it, followed by wavelet threshold denoising to obtain a denoised signal; processing the denoised one-dimensional long-time series signal by sliding slices and inputting it into a pre-constructed partial discharge identification model for training; inputting the target signal into the trained partial discharge identification model to obtain the time location and duration of the actual partial discharge event in the long-time series signal, thereby achieving effective identification and separation of partial discharge signals and corona interference in cable terminals of new energy power plants. This invention integrates wavelet denoising, time windowing, and one-dimensional convolutional network deep feature recognition, enabling stable identification of partial discharge signals in high-noise and mixed-pulse scenarios in cable terminals of new energy power plants, providing a reliable data foundation for insulation status assessment and maintenance early warning of cable terminals in new energy power plants.

Description

Technical Field

[0001] This invention relates to the field of online monitoring and intelligent diagnosis technology for the insulation status of high-voltage cables in new energy power stations, and in particular to a method and system for identifying partial discharge signals at cable terminals in new energy power stations. Background Technology

[0002] Cable terminals in new energy power plants (photovoltaic, wind power, energy storage, etc.) are used to connect key high-voltage components such as combiner boxes, inverters, and main transformers. They are mostly installed outdoors or in distribution rooms and are susceptible to extreme weather, vibration, and electromagnetic interference within the plant. During manufacturing or long-term operation, defects such as air gaps and interface debonding can easily develop inside the terminals, leading to partial discharge under voltage. If these defects are not identified and located in a timely manner, they can cause insulation deterioration, threatening the safe operation of the power plant. In existing technologies, detection often employs high-frequency current methods or ultra-high-frequency methods. However, due to numerous interference sources at the site (corona discharge, switching actions, inverter pulses, etc.), the original signal exhibits low signal-to-noise ratio and a long-term mixed waveform. Traditional wavelet denoising and simple feature + classifier methods struggle to provide stable identification in scenarios with pulse truncation and similar amplitudes. Existing wavelet transform denoising techniques, while addressing the issues of "high noise and difficulty in threshold selection," lack a complete solution—they cannot automatically identify discharge types or reconstruct long-term mixed signals. This invention builds upon these techniques by adding an adaptive time window and a one-dimensional convolutional neural network module to adapt to the field environment and meet the needs of online detection and real-time identification. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for identifying partial discharge signals in cable terminals of new energy power plants. Through an integrated process of "acquisition → wavelet threshold noise reduction → time window slicing → one-dimensional convolutional neural network recognition → time axis reconstruction", the invention achieves automatic identification and separation of partial discharge events in long-sequence mixed signals of cable terminals of new energy power plants, thereby improving the accuracy and real-time performance of insulation status assessment of cable terminals of new energy power plants.

[0004] To achieve the above objectives, the present invention provides the following technical solution: On one hand, the present invention provides a method for identifying partial discharge signals at cable terminals in new energy power stations, comprising the following steps: S1. Collect the original partial discharge signal of the cable terminal of the new energy power station, and perform preprocessing such as DC removal and normalization to ensure that the signal meets the zero mean condition. S2. The original signal is denoised by a preset wavelet basis and decomposition level to obtain a denoised one-dimensional signal sequence. S3. The denoised one-dimensional signal sequence is sliced ​​according to the preset time window length and step size to obtain several one-dimensional signal samples for deep learning recognition, and input into the constructed partial discharge recognition model based on one-dimensional convolutional neural network for training. S4. Input the processed target signal into the trained partial discharge recognition model. The trained partial discharge recognition model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are identified as partial discharge, and obtains the time location and duration of partial discharge events in the long time series signal, thus completing the signal recognition.

[0005] On the other hand, the present invention provides a partial discharge signal identification system for cable terminals in new energy power stations, which uses the above method and includes the following modules: Signal acquisition module: used to acquire and preprocess the raw partial discharge signals of cable terminals in new energy power plants; Noise reduction module: Denoises the original signal to obtain a denoised one-dimensional signal sequence; Slicing and training module: The denoised one-dimensional signal sequence is sliced ​​by sliding to obtain several one-dimensional signal samples for deep learning recognition, and then input into the constructed partial discharge recognition model based on a one-dimensional convolutional neural network for training. The identification output module inputs the processed target signal into the trained partial discharge identification model. The trained partial discharge identification model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are identified as partial discharge, and obtains the time location and duration of partial discharge events in the long time series signal, thus completing the signal identification.

[0006] Compared with the prior art, the present invention has the following beneficial effects: This invention directly connects to a one-dimensional convolutional neural network classification module after the wavelet threshold denoising stage. This overcomes the limitations of traditional techniques that can only output "purified waveforms," ​​directly generating clear labels such as "partial discharge," "corona interference," and "mixed signal," enabling qualitative identification of the detected signal and significantly improving the intuitiveness and application value of signal analysis. Through time window sliding and time axis reconstruction techniques, it can effectively separate multiple discharge and interference events that occur intermittently in long-sequence mixed signals, accurately capturing irregular pulse signal characteristics, perfectly adapting to the application scenarios of new energy power station cable terminals, which face numerous environmental interferences and irregular pulse occurrences. The network training uses the Adam optimization algorithm and cross-entropy loss function, combined with the improved signal-to-noise ratio from the wavelet preprocessing stage, forming a "preprocessing-training" collaborative mechanism. The entire process uses one-dimensional signals as the processing object, eliminating the need for additional time-frequency diagrams or large-size images, significantly reducing data processing and model computation. Its lightweight design allows for easy deployment on industrial control computers or edge computing units, adapting to scenarios with limited hardware computing power and enabling real-time online detection. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall process of Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of wavelet threshold noise reduction processing in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the network structure of the partial discharge identification model in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the noise reduction result of the wavelet function in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the reconstruction on the time axis after long-sequence mixed signal identification in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the overall structure of Embodiment 2 of the present invention. Detailed Implementation

[0008] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0009] Example 1: Please see Figures 1-5 A method for identifying partial discharge signals at cable terminals in new energy power stations includes the following steps: S1. Use a high-frequency current sensor installed at the grounding wire or shielding layer of the cable terminal in the new energy power station to collect the original one-dimensional waveform signal containing partial discharge signal and corona interference signal, and perform preprocessing, including removing DC component and normalization, to ensure that the input signal meets the zero mean condition: ; In the formula, This is the original signal of partial discharge at the cable terminal; The mean; Standard deviation; This is a one-dimensional signal obtained by removing the mean from the original signal and normalizing it according to the standard deviation.

[0010] S2. The original signal is denoised to obtain a denoised one-dimensional signal sequence.

[0011] The signal is decomposed using one of the following wavelet bases: Symlet-4 wavelet, Coiflets wavelet, or Daubechies wavelet.

[0012] In one specific embodiment, firstly, a Symlet-4 wavelet basis with good orthogonality and approximate symmetry is selected, and the signal is decomposed into five levels to obtain approximate coefficients. and 5 layers of detail coefficient This process achieves the separation of white noise and partial discharge signals in the time and frequency domains.

[0013] ; In the formula, These are approximate coefficients; For the first Layer detail factor; The number of floors.

[0014] Low frequency part The overall signal profile was preserved; the high-frequency portion It mainly reflects high-frequency noise; mid- and low-frequency components. It contains characteristics of partial discharge.

[0015] Then, the noise variance for each layer is calculated using the empirical Bayesian method: ; For each layer of high-frequency wavelet coefficients, noise variance is estimated, and the variance of the noisy signal is estimated as follows: ; ; In the formula, The standard deviation of the noisy signal; The standard deviation of the pure signal components; The standard deviation of the noise component; For the first Layer wavelet coefficients The median of the absolute values; 0.6745 is the adjustment factor for the standard deviation of Gaussian white noise; This represents the length of each dimension of the wavelet coefficients in the current layer. Number of floors; This is the index for the layer coefficient position; This is a column for layer coefficients.

[0016] Apply a semi-soft thresholding function to the detail coefficients of each layer: ; In the formula, For the first Layer wavelet coefficients; This is the lower threshold of the semi-soft thresholding function; This is the upper threshold of the semi-soft thresholding function; It is a symbolic function.

[0017] The processed detail coefficients are reconstructed using inverse wavelet transform to obtain the denoised signal. For example... Figure 4 As shown, the experimental results indicate that the signal noise is significantly reduced and the partial discharge pulse is clearly visible.

[0018] S3. The denoised one-dimensional signal sequence is sliced ​​by sliding to obtain several one-dimensional signal samples for deep learning recognition. The signal samples are divided into training set and validation set according to an 8:2 ratio and input into the constructed partial discharge recognition model based on one-dimensional convolutional neural network (1DCNN) for training.

[0019] By using a sliding time window, long-sequential, intermittent, and mixed pulse sequences are broken down into multiple short sequences, enabling them to be efficiently recognized by a one-dimensional convolutional neural network and reconstructed as specific discharge events on the time axis. Specifically, the denoised signal is sliced ​​into 1024-point segments with a step size of 512 points, with each segment serving as a one-dimensional sample.

[0020] The partial discharge identification model comprises an input layer, at least two convolutional layers and pooling layers, a batch normalization layer, a nonlinear activation layer, a fully connected layer, and a Softmax output layer, connected in sequence. The convolutional kernels in the convolutional layers have a kernel length of 3, and the number of kernels increases progressively with each layer. The Softmax output layer outputs the probabilities of the partial discharge, corona / pulse interference, and mixed discharge categories. Specific parameters are as follows: Table 1 Parameters of Partial Discharge Identification Model

[0021] During model training, the Adam optimizer and cross-entropy loss are employed to enhance convergence performance under conditions of insufficient sample size or imbalanced class composition. The Adam algorithm is an adaptive learning rate optimization algorithm that combines first-moment and second-moment estimation of the gradient, adjusting the learning rate by calculating the exponential moving average and squared exponential moving average of the gradient. The implementation process is as follows: ; ; ; ; ; in, ; In the formula, For the first First-moment estimate of the gradient at step time (the exponential moving average of the gradient, i.e., "momentum"); This represents the number of iteration steps. The attenuation rate is estimated by the first moment; For the first First-moment estimation of the gradient at each step; For the first gradient at step; For the first Second-moment estimate of the gradient at step time (exponential moving average of the squared gradient). The attenuation rate is estimated by the second moment; This is the first-order moment estimate after bias correction; , All are attenuation rates Power; This is the second-order moment estimate after bias correction; For the first Model parameters at step time; The learning rate (step size); is a small constant used for numerical stability.

[0022] The training process employed a mini-batch stochastic gradient descent strategy with a batch size of 64 and 500 epochs of iterative training. Other parameter settings are shown in Table 2. The loss and accuracy of the training and validation sets were monitored. Training was terminated when the accuracy of the validation set converged and overfitting was under control. The model accuracy was 98.5185%.

[0023] Table 2 Training Parameter Settings

[0024] S4. Input the target long-time series signal into the trained partial discharge recognition model. The trained partial discharge recognition model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are judged as partial discharge, obtains the time location and duration of partial discharge events in the long-time series signal, and removes or archives the time windows that are judged as corona interference.

[0025] For the target long-time series signal collected from the cable terminal of the new energy power station within a certain time period, the above wavelet denoising and time window slicing process is repeated to obtain a series of window samples arranged in chronological order. Then, a trained partial discharge recognition model is used for recognition and prediction. Figure 5 As shown, the partial discharge, corona, and mixed signals identified will be merged on the time axis to obtain a partial discharge event at the cable terminal of the new energy power station.

[0026] Example 2 A partial discharge signal identification system for cable terminals in new energy power stations, using the above-mentioned method, such as... Figure 6 As shown, it includes the following modules: Signal acquisition module: used to acquire and preprocess the raw partial discharge signals of cable terminals in new energy power plants; Noise reduction module: Denoises the original signal to obtain a denoised one-dimensional signal sequence; Slicing and training module: The denoised one-dimensional signal sequence is sliced ​​by sliding to obtain several one-dimensional signal samples for deep learning recognition, and then input into the constructed partial discharge recognition model based on a one-dimensional convolutional neural network for training. The identification output module inputs the target signal, which has also been processed by the above module, into the trained partial discharge identification model. The trained partial discharge identification model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are identified as partial discharge, and obtains the time location and duration of the partial discharge event in the long time series signal, thus completing the signal identification.

[0027] This invention discloses a partial discharge signal identification system for cable terminals in new energy power stations, which can be installed in a computer device. The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a partial discharge signal identification program for cable terminals in new energy power stations. The memory includes at least one type of readable storage medium, including flash memory, portable hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic storage, disk, optical disk, etc. The processor is the control core of the electronic device, connecting various components of the computer device via various interfaces and lines. It executes programs or modules stored in the memory and calls data stored in the memory to perform various functions of the computer device and process data.

[0028] The module described in this invention refers to a series of computer program segments that can be executed by the processor of a computer device and can perform a fixed function, and which are stored in the memory of the computer device.

[0029] The system provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0030] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall fall within the scope of the present invention.

Claims

1. A method for identifying partial discharge signals at cable terminals in new energy power stations, characterized in that, Includes the following steps: S1. Collect and preprocess the original partial discharge signal of the cable terminal of the new energy power station. S2. Perform wavelet threshold denoising on the original signal using a preset wavelet basis and decomposition level to obtain a denoised one-dimensional signal sequence; the wavelet basis includes one of Symlet-4 wavelet, Coiflets wavelet or Daubechies wavelet, and the decomposition level is 3-5 levels. The noise variance for each layer was calculated using the empirical Bayesian method. ； For each layer of high-frequency wavelet coefficients, noise variance is estimated, and the variance of the noisy signal is estimated as follows: ； ； In the formula, The standard deviation of the noisy signal; The standard deviation of the pure signal components; The standard deviation of the noise component; For the first Layer wavelet coefficients The median of the absolute values; 0.6745 is the adjustment factor for the standard deviation of Gaussian white noise; This represents the length of each dimension of the wavelet coefficients in the current layer. Number of floors; This is the index for the layer coefficient position; Columns for layer coefficients; The signal was decomposed into five levels using Symlet-4 wavelets to obtain approximate coefficients. and 5 layers of detail coefficient The processed detail coefficients are reconstructed using inverse wavelet transform to obtain the denoised signal; a semi-soft thresholding function is applied to the detail coefficients. ； In the formula, For the first Layer wavelet coefficients; This is the lower threshold of the semi-soft thresholding function; This is the upper threshold of the semi-soft thresholding function; It is a symbolic function; S3. The denoised one-dimensional signal sequence is sliced ​​by sliding to obtain several one-dimensional signal samples for deep learning recognition, and then input into the constructed partial discharge recognition model based on a one-dimensional convolutional neural network for training. S4. Input the processed target signal into the trained partial discharge recognition model. The trained partial discharge recognition model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are identified as partial discharge, and obtains the time location and duration of partial discharge events in the long time series signal, thus completing the signal recognition.

2. The method for identifying partial discharge signals at cable terminals in new energy power stations according to claim 1, characterized in that, In S1, the preprocessing includes DC removal and normalization operations.

3. The method for identifying partial discharge signals at cable terminals in new energy power stations according to claim 1, characterized in that, In S4, the length of the time window is 256-1024 sampling points, and the step size is 1 / 4-1 / 2 of the length of the time window.

4. The method for identifying partial discharge signals at cable terminals in new energy power stations according to claim 1, characterized in that, The partial discharge identification model includes an input layer, at least two levels of convolutional layers and pooling layers, a batch normalization layer, a nonlinear activation layer, a fully connected layer, and a Softmax output layer connected in sequence. The convolutional kernels of the convolutional layers have a length of 3, and the number of convolutional kernels increases layer by layer. The Softmax output layer is used to output the probabilities of the partial discharge class, corona / pulse interference class, and mixed class.

5. A partial discharge signal identification system for cable terminals in new energy power stations, using the method described in any one of claims 1-4, comprising: Signal acquisition module: used to acquire and preprocess the raw partial discharge signals of cable terminals in new energy power plants; Noise Reduction Module: Performs wavelet threshold noise reduction on the original signal using a preset wavelet basis and decomposition level to obtain a noise-reduced one-dimensional signal sequence; the wavelet basis includes one of Symlet-4 wavelet, Coiflets wavelet, or Daubechies wavelet, and the decomposition level is 3-5 levels; The noise variance for each layer was calculated using the empirical Bayesian method. ； For each layer of high-frequency wavelet coefficients, noise variance is estimated, and the variance of the noisy signal is estimated as follows: ； ； In the formula, The standard deviation of the noisy signal; The standard deviation of the pure signal components; The standard deviation of the noise component; For the first Layer wavelet coefficients The median of the absolute values; 0.6745 is the adjustment factor for the standard deviation of Gaussian white noise; This represents the length of each dimension of the wavelet coefficients in the current layer. Number of floors; This is the index for the layer coefficient position; Columns for layer coefficients; The signal was decomposed into five levels using Symlet-4 wavelets to obtain approximate coefficients. and 5 layers of detail coefficient The processed detail coefficients are reconstructed using inverse wavelet transform to obtain the denoised signal; a semi-soft thresholding function is applied to the detail coefficients. ； In the formula, For the first Layer wavelet coefficients; This is the lower threshold of the semi-soft thresholding function; This is the upper threshold of the semi-soft thresholding function; It is a symbolic function; Slicing and training module: The denoised one-dimensional signal sequence is sliced ​​by sliding to obtain several one-dimensional signal samples for deep learning recognition, and then input into the constructed partial discharge recognition model based on a one-dimensional convolutional neural network for training. The identification output module inputs the processed target signal into the trained partial discharge identification model. The trained partial discharge identification model reconstructs the classification results of each time window arranged in chronological order, merges the time windows that are identified as partial discharge, and obtains the time location and duration of partial discharge events in the long time series signal, thus completing the signal identification.

Images