Wired insulation on-line monitoring method and system based on digital intelligent power supply storage

Abstract

The invention discloses a wired insulation on-line monitoring method and system based on digital intelligent power supply storage, and the method comprises the steps: firstly collecting and preprocessing a power interference signal, extracting feature information from the signal through a wavelet transform technology, and then carrying out the classification and prediction of a cable insulation state through a Transform model. The method not only can effectively identify the health state, the slight aging state and the severe aging state of the cable, but also is high in detection precision and good in real-time performance, overcomes the problems that an existing monitoring method is insufficient in real-time performance and low in detection precision in a complex electric power environment, and remarkably improves the reliability and efficiency of cable insulation monitoring.

Description

Technical Field

[0001] The invention relates to a wired insulation online monitoring method and system based on digital intelligent power supply storage, belonging to the technical field of insulation monitoring. Background Art

[0002] Cables are an important part of the power system, and their insulation status directly affects the stability and safety of the power system. As the service life of cables increases, insulation aging or damage problems gradually emerge, resulting in frequent power system failures, posing a huge challenge to power supply stability and safety. Traditional cable insulation monitoring methods have problems such as poor real-time performance and low fault identification accuracy, and are unable to detect changes in insulation status in a timely and effective manner. Therefore, in-depth research on cable insulation online monitoring technology, especially the introduction of digital intelligence and machine learning technology, will have a profound impact on the development of cable detection technology. Summary of the invention

[0003] The purpose of the present invention is to provide a wired insulation online monitoring method and system based on digital intelligent power storage, so as to solve the problems of poor real-time performance and low fault identification accuracy in existing traditional cable insulation monitoring methods.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A wired insulation online monitoring method based on digital intelligent power storage is composed of an input layer, a multi-head self-attention layer, a feedforward neural network and an output layer to form a Transformer model, including the following steps:

[0006] S1: Collect interference signals. The interference signals are collected through high-precision current sensors and high-precision voltage sensors; S2: Preprocess the interference signals to form feature vectors; S3: Extract features from feature vectors to form fault signal features; S4: Process the fault signal features through the input layer to form an embedding vector; S5: Process the embedding vector through a multi-head self-attention layer to form a multi-head vector; S6: Process the multi-head vector through a feedforward neural network to form a discriminant vector; S7: Process the discriminant vector through the output layer to obtain the insulation status of the cable.

[0007] The purpose of the present invention can also be further achieved by the following technical measures:

[0008] The aforementioned wired insulation online monitoring method based on digital intelligent power storage preferably comprises the following steps in step S2:

[0009] S21. The interference signal is processed by a bandpass filter to form a concentrated signal. The transfer function of the bandpass filter is as follows:

[0010]

[0011] Where: f L and f H Represent the lower and upper frequencies of the bandpass filter respectively;

[0012] S22, performing denoising on the concentrated signal by wavelet transform to form a preliminary feature vector;

[0013] S23. Process the preliminary feature vector by a soft threshold method to form a feature vector.

[0014] In the aforementioned wired insulation online monitoring method based on digital intelligent power storage, preferably, the processing method of the soft threshold method in the above step S23 is:

[0015]

[0016] Where sgn(x) represents the sign function of x, σ represents the standard deviation of the preliminary eigenvector, and N represents the length of the signal.

[0017] In the aforementioned wired insulation online monitoring method based on digital intelligent power storage, preferably, the processing step in the above step S3 specifically includes: decomposing the feature vector by wavelet transform, and the decomposition of the wavelet transform is as follows:

[0018]

[0019] Where: A k Indicates the jth 0 Approximation coefficient of the layer, D jk represents the detailed coefficient of the jth layer, represents the scaling function, φ jk (t represents the wavelet function.

[0020] In the aforementioned wired insulation online monitoring method based on digital intelligent power storage, preferably, the attention mechanism formula in the above step S5 is as follows:

[0021]

[0022] Where: Q represents the query matrix, K represents the key matrix, V represents the value matrix, d k represents the dimension of the key matrix, and softmax() represents average pooling.

[0023] In the aforementioned wired insulation online monitoring method based on digital intelligent power storage, preferably, the above step S56 processes the multi-head vector through a multi-layer feedforward neural network to form a discriminant vector.

[0024] In the aforementioned wired insulation online monitoring method based on digital intelligent power storage, preferably, the loss function of the Transformer model is as follows:

[0025]

[0026] Where: C represents the number of categories, y i represents the true label of the i-th category, Represents the probability value of the i-th category predicted by the model.

[0027] Based on another purpose of the present invention, the present invention provides the following technical solutions: a wired insulation online monitoring system based on digital intelligent power storage, the system uses the monitoring method as mentioned above, including: an acquisition module for acquiring interference signals; a preprocessing module for preprocessing the interference signals to form a feature vector; an extraction module for extracting features from the feature vector to form a fault signal feature; an input module for processing the fault signal feature to form an embedded vector; a multi-head self-attention module for processing the embedded vector to form a multi-head vector; a feedforward neural network module for processing the multi-head vector to form a discriminant vector; and an output module for processing the discriminant vector to obtain the insulation state of the cable.

[0028] The present invention also provides the following technical solution: a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the wired insulation online monitoring method based on digital intelligent power supply storage as described above is implemented.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention first collects and preprocesses the power interference signal, extracts characteristic information from the signal using wavelet transform technology, and then classifies and predicts the cable insulation status through the Transformer model. This method can not only effectively identify the healthy state, slight aging and severe aging state of the cable, but also has high detection accuracy and good real-time performance, overcoming the problems of insufficient real-time performance and low detection accuracy of the existing monitoring methods in complex power environments, and significantly improving the reliability and efficiency of cable insulation monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic flow chart of a wired insulation online monitoring method based on digital intelligent power storage;

[0031] Figure 2 It is a schematic diagram of the system structure of the present invention.

[0032] In the figure: 100, monitoring system; 110, acquisition module; 120, preprocessing module; 130, extraction module; 140, input module; 150, multi-head self-attention module; 160, feedforward neural network module; 170, output module. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] Embodiment 1

[0035] like Figure 1 As shown, a wired insulation online monitoring method based on digital intelligent power storage includes the following steps:

[0036] S1: Collect interference signals. Interference signals are collected by high-precision current sensors and high-precision voltage sensors. The interference signal collection step essentially includes three parts, namely real-time signal collection, preliminary signal filtering and processing, and signal storage. In the real-time collection process, high-precision current sensors and high-precision voltage sensors are used for collection. The sampling frequency is 10kHz, wherein the accuracy of the voltage sensor is preferably less than 0.1%, and the measurement range of the high-precision current sensor is 0-1000 (A). A high-bandwidth data acquisition card is used in the collection process, and signals can be continuously and stably acquired under different operating environments. The bandwidth of the preferred data acquisition card is 1.5MHz.

[0037] S2: Preprocess the interference signal to form a feature vector; the preprocessing process of the interference signal includes filtering and denoising. Since the power interference signal contains a lot of noise and interference, directly analyzing the original signal may lead to errors, especially when identifying the insulation status of the cable. Therefore, preprocessing is a key step in signal analysis, and its main purpose is to remove irrelevant noise and retain the characteristic signal related to the insulation status.

[0038] S3: Extract features from feature vectors to form fault signal features; feature extraction mainly uses wavelet transform, which is a time-frequency analysis method that decomposes the signal into wavelet functions of different frequency ranges and analyzes the local characteristics of the signal. Unlike traditional Fourier transform, wavelet transform can perform time domain and frequency domain analysis at different scales, and is particularly suitable for processing transient and high-frequency fault signals. The basic principle is to decompose the original signal layer by layer by scaling and translating the mother wavelet function to capture the local characteristics of the signal at multiple scales. In cable insulation monitoring, the signal usually includes low-frequency normal operation signals and high-frequency fault signals. Wavelet transform can separate these two types of signals, so that the fault signal is amplified and highlighted in the high-frequency component, which is convenient for subsequent analysis.

[0039] S4: Process the fault signal features through the input layer to form an embedding vector;

[0040] S5: The embedding vector is processed through a multi-head self-attention layer to form a multi-head vector;

[0041] S6: Process the multi-head vector through a feedforward neural network to form a discriminant vector;

[0042] S7: The discriminant vector is processed through the output layer to obtain the insulation state of the cable. The input layer, multi-head self-attention layer, feedforward neural network and output layer constitute the Transformer model. The core of the Transformer model is the self-attention mechanism, which can effectively capture the long-distance dependencies between elements in sequence data. Compared with traditional recurrent neural networks or long short-term memory networks, the Transformer model has parallel computing capabilities and shows higher training efficiency and accuracy when processing large-scale time series data. In the insulation state recognition task, the power interference signal generated by the cable operation can be regarded as time series data. The Transformer model can capture the complex features related to the insulation state change.

[0043] Specifically, the present invention first collects and preprocesses the power interference signal, extracts characteristic information from the signal using wavelet transform technology, and then classifies and predicts the cable insulation status through the Transformer model. This method can not only effectively identify the healthy state, slight aging and severe aging state of the cable, but also has high detection accuracy and good real-time performance, overcoming the problems of insufficient real-time performance and low detection accuracy of existing monitoring methods in complex power environments, and significantly improving the reliability and efficiency of cable insulation monitoring.

[0044] Preferably, the above S2 specifically includes the following steps:

[0045] S21. The interference signal is processed by a bandpass filter to form a concentrated signal. The transfer function of the bandpass filter is as follows:

[0046]

[0047] Where: f L and f H Represent the lower and upper limits of the bandpass filter, respectively. The preprocessing first uses a bandpass filter to remove low-frequency power frequency interference and high-frequency spike noise from the signal. The power frequency interference in the power system is mainly concentrated at 50 Hz, so the passband frequency of the filter is set between 100 Hz and 1 kHz, which can effectively remove the power frequency interference and retain the high-frequency interference signal caused by insulation faults;

[0048] S22, performing denoising on the concentrated signal by wavelet transform to form a preliminary feature vector. The formation process of the preliminary feature vector is as follows:

[0049] x(t)=s(t)+n(t)

[0050] In the formula, s(t) represents the original signal, n(t) represents the noise signal. Through wavelet transform, the signal is decomposed into wavelet coefficients of different frequency scales to form a preliminary feature vector. In the wavelet denoising process, the Dobex wavelet (db4) is selected as the mother wavelet. The high-frequency noise component is identified and filtered out through the multi-scale decomposition of the signal, while retaining the disturbance characteristics closely related to the aging and damage of the cable insulation.

[0051] S23. Processing the preliminary feature vector by a soft threshold method to form a feature vector.

[0052] Preferably, the processing method of the soft threshold method in S23 is:

[0053]

[0054] Where: sgn(x) represents the sign function of x, σ represents the standard deviation of the preliminary feature vector, N represents the length of the signal, when x>0, sgn(x)=1, when x=0, sgn(x)=0, when x<0, sgn(x)=-1. The purpose of interference signal acquisition and preprocessing is to extract feature data closely related to the insulation state of the cable, laying the foundation for subsequent machine learning model analysis. Through high-precision sensor configuration and optimized signal processing technology, this method can effectively extract key interference features, thereby improving the fault identification accuracy and ultra-full stability of the method.

[0055] Furthermore, the feature extraction process mainly uses wavelet transform processing. Wavelet transform is a time-frequency analysis method that decomposes the signal into wavelet functions of different frequency ranges and analyzes the local characteristics of the signal. Unlike the traditional Fourier transform, wavelet transform can perform time domain and frequency domain analysis at different scales at the same time, which is particularly suitable for processing transient and high-frequency fault signals. The basic principle is to decompose the original signal layer by layer by scaling and translating the mother wavelet function to capture the local characteristics of the signal at multiple scales. In cable insulation monitoring, the signal usually includes low-frequency normal operating signals and high-frequency fault signals. Wavelet transform can separate these two types of signals so that the fault signal is amplified and highlighted in the high-frequency component, which is convenient for subsequent analysis. The feature vector is decomposed by wavelet transform, and the decomposition of wavelet transform is as follows:

[0056]

[0057] Where: A k Indicates the jth0 Approximation coefficient of the layer, D jk represents the detailed coefficient of the jth layer, represents the scaling function, φ jk (t) represents the wavelet function. Multi-scale decomposition can capture fault characteristics in different frequency ranges. For example, faults such as partial discharge and insulation damage usually cause high-frequency anomalies. By analyzing these detail coefficients, potential faults in cables can be effectively identified.

[0058] The above wavelet transform process can separate signals with different frequency components from the original power interference signal, thereby more accurately extracting the fault signal characteristics. The approximate coefficients and detail coefficients obtained by wavelet decomposition can effectively reflect the health status of cable operation and potential fault characteristics.

[0059] Furthermore, the processing steps in the above S5 are as follows:

[0060]

[0061] Where: Q represents the query matrix, K represents the key matrix, V represents the value matrix, d k Represents the dimension of the key matrix. The multi-head self-attention layer uses the multi-head self-attention mechanism to process the embedding vector, so that the model can process different signal features in parallel and combine the local and global features in the signal.

[0062] Furthermore, the above S6 specifically includes the following steps: the multi-head vector is processed by a multi-layer feedforward neural network to form a discriminant vector. A feedforward neural network is a basic neural network structure, which is characterized by a one-way flow of information from the input layer through the hidden layer (if any) to the output layer, and there is no feedback loop. A feedforward neural network usually includes an input layer, one or more hidden layers, and an output layer, in which neurons are fully connected, that is, each neuron is connected to all neurons in the previous layer. In a feedforward neural network, neurons in the hidden layer usually use activation functions to introduce nonlinearity so that the network can learn and simulate complex input-output relationships. The working principle of a feedforward neural network includes two processes: forward propagation and back propagation. During the forward propagation process, the network calculates the output based on the current weights; during the back propagation process, the network calculates the gradient based on the error between the output and the target, and uses these gradients to update the weights and biases of the network to reduce the error. This training process is usually repeated many times until the performance of the network reaches a satisfactory level. Through a multi-layer feedforward neural network, the nonlinear expression ability of the model is further improved.

[0063] Furthermore, the loss function of the Transformer model is as follows:

[0064]

[0065] Where: C represents the number of categories, y i represents the true label of the i-th category, Represents the probability value of the i-th category predicted by the model.

[0066] The Transformer model also uses the following optimization strategies to improve the performance of the model:

[0067] Learning rate scheduler: In order to avoid overfitting during training, the present invention adopts a strategy of dynamically adjusting the learning rate during training, so that the model converges quickly with a higher learning rate in the early stage of learning, and fine-tunes with a lower learning rate in the later stage of training to ensure the stability of the model.

[0068] Regularization techniques: including stopping learning and weight decay, can effectively reduce model overfitting and improve the model's generalization ability for test data.

[0069] Early stopping method: By verifying the performance indicators, if the accuracy of the validation set no longer improves after multiple consecutive trainings, the training is stopped early to avoid performance degradation caused by overtraining of the model.

[0070] The applicant needs to emphasize that the input sequence of the model not only contains the numerical value of the voltage or current disturbance signal, but also encodes the position information of each time point into the input vector through position encoding, so that the model can identify the signal characteristics at different positions in the time domain. This design enables the model to better capture the potential local and global characteristics in the cable signal, especially the occurrence and propagation of fault signals.

[0071] Embodiment 2

[0072] The same parts in this embodiment as those in the first embodiment are given the same figure marks, and the same text descriptions are omitted.

[0073] like Figure 2 As shown, this embodiment discloses a wired insulation online monitoring system based on digital intelligent power storage. The monitoring system 10 uses the monitoring method in Example 1, including: an acquisition module 110, used to acquire interference signals; a preprocessing module 120, used to preprocess the interference signals to form a feature vector; an extraction module 130, used to extract features from the feature vector to form a fault signal feature; an input module 140, used to process the fault signal feature to form an embedded vector; a multi-head self-attention module 150, used to process the embedded vector to form a multi-head vector; a feedforward neural network module 160, used to process the multi-head vector to form a discriminant vector; an output module 170, used to process the discriminant vector to obtain the insulation state of the cable.

[0074] Embodiment 3

[0075] The same parts as those in the first and second embodiments of the present embodiment are given the same reference numerals, and the same text descriptions are omitted.

[0076] This embodiment discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the monitoring method in the first embodiment is implemented.

[0077] In addition to the above embodiments, the present invention may also have other implementation modes. Any technical solutions formed by equivalent replacement or equivalent transformation shall fall within the protection scope required by the present invention.

Claims

1. A wired insulation online monitoring method based on digital intelligent power storage, characterized in that: The Transformer model consists of an input layer, a multi-head self-attention layer, a feedforward neural network, and an output layer, and includes the following steps: S1: Collect interference signals. The interference signals are collected through high-precision current sensors and high-precision voltage sensors; S2: Preprocess the interference signals to form feature vectors; S3: Extract features from feature vectors to form fault signal features; S4: Process the fault signal features through the input layer to form an embedding vector; S5: Process the embedding vector through a multi-head self-attention layer to form a multi-head vector; S6: Process the multi-head vector through a feedforward neural network to form a discriminant vector; S7: Process the discriminant vector through the output layer to obtain the insulation status of the cable.

2. The wired insulation online monitoring method based on digital intelligent power storage according to claim 1 is characterized in that: Specifically in step S2 The following steps are involved: S21. The interference signal is processed by a bandpass filter to form a concentrated signal. The transfer function of the bandpass filter is as follows: Where: f L and f H Represent the lower and upper frequencies of the bandpass filter respectively; S22, performing denoising on the concentrated signal by wavelet transform to form a preliminary feature vector; S23. Process the preliminary feature vector by a soft threshold method to form a feature vector.

3. The wired insulation online monitoring method based on digital intelligent power storage according to claim 2 is characterized in that: The processing method of the soft threshold method in step S23 is: Where sgn(x) represents the sign function of x, σ represents the standard deviation of the preliminary eigenvector, and N represents the length of the signal.

4. The wired insulation online monitoring method based on digital intelligent power storage according to claim 1 is characterized in that: The processing steps in step S3 specifically include: decomposing the feature vector by wavelet transform, and the decomposition of wavelet transform is as follows: Where: A k represents the approximation coefficient of the j0th layer, D jk represents the detailed coefficient of the jth layer, represents the scaling function, φ jk (t) represents the wavelet function.

5. The wired insulation online monitoring method based on digital intelligent power storage according to claim 1 is characterized in that: The attention mechanism formula in step S5 is as follows: Where: Q represents the query matrix, K represents the key matrix, V represents the value matrix, d k represents the dimension of the key matrix, and softmax() represents average pooling.

6. The wired insulation online monitoring method based on digital intelligent power storage according to claim 1 is characterized in that: Step S6 processes the multi-head vector through a multi-layer feedforward neural network to form a discriminant vector.

7. The wired insulation online monitoring method based on digital intelligent power storage according to claim 1 is characterized in that: The loss function of the Transformer model is as follows: Where: C represents the number of categories, y i represents the true label of the i-th category, Represents the probability value of the i-th category predicted by the model.

8. A wired insulation online monitoring system based on digital intelligent power storage, the system uses the wired insulation online monitoring method based on digital intelligent power storage as described in claims 1-7, characterized in that: The system includes: an acquisition module for acquiring interference signals; a preprocessing module for preprocessing interference signals to form feature vectors; an extraction module for extracting features from the feature vectors to form fault signal features; an input module for processing the fault signal features to form embedded vectors; a multi-head self-attention module for processing the embedded vectors to form multi-head vectors; a feedforward neural network module for processing the multi-head vectors to form discriminant vectors; and an output module for processing the discriminant vectors to obtain the insulation status of the cable.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the wired insulation online monitoring method based on digital intelligent power supply storage as described in claims 1-7 is implemented.

Images