Monitoring protection strategy method for vehicle-mounted solid insulation high-voltage system

Abstract

The invention provides a monitoring protection strategy method for a vehicle-mounted solid insulation high-voltage system, and relates to the technical field of intelligent monitoring and fault diagnosis of high-voltage equipment, and the method comprises the steps: carrying out the real-time collection of an ultrasonic voiceprint signal generated in the operation process of the solid insulation high-voltage system through a voiceprint collection device; processing through a blind source separation algorithm to obtain a partial discharge voiceprint signal of the solid insulation high-voltage system; constructing a partial discharge voiceprint signal database of the solid insulation high-voltage system; constructing a first partial discharge voiceprint diagnosis model by using a gated recurrent neural network as a core to obtain an optimized second partial discharge voiceprint diagnosis model; and applying the trained second partial discharge voiceprint diagnosis model to online monitoring of the solid insulation high-voltage system, comparing the extracted signal features with the partial discharge signal features in the database, and judging whether the extracted signal features are consistent with the partial discharge signal features. The method has the beneficial effects that the partial discharge signal in the operation of the solid insulation high-voltage system can be efficiently and accurately identified, and the reliability and the safety of the operation of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and fault diagnosis of high-voltage equipment. Specifically, it relates to a monitoring and protection strategy method for an in-vehicle solid-insulated high-voltage system. Background Art

[0002] An in-vehicle solid-insulated high-voltage system is a system in which a vacuum circuit breaker, a high-voltage disconnector, a high-voltage voltage transformer, an earthing switch, and a group of lightning arresters are all assembled in a closed steel box, and solid insulation material is injected into the iron box. Compared with traditional gas high-voltage boxes, the solid high-voltage system has the advantages of smaller volume, more compact structure, lower maintenance cost, environmental friendliness, higher safety and reliability, lower noise, and longer life. However, the online monitoring of traditional solid-insulated high-voltage systems has problems such as slow data transmission and processing speed and low accuracy, which affect the ability of real-time monitoring and rapid fault response. Summary of the Invention

[0003] The purpose of the present invention is to provide a monitoring and protection strategy method for an in-vehicle solid-insulated high-voltage system to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0004] In a first aspect, the present application provides a monitoring and protection strategy method for an in-vehicle solid-insulated high-voltage system, including:

[0005] Using a soundprint acquisition device to collect ultrasonic soundprint signals generated during the operation of the solid-insulated high-voltage system in real time to obtain original soundprint data containing partial discharge information;

[0006] Based on the original soundprint data, through blind source separation algorithm processing, the partial discharge soundprint signal of the solid-insulated high-voltage system is obtained;

[0007] According to the obtained partial discharge soundprint signal, a partial discharge soundprint signal database of the solid-insulated high-voltage system is constructed, where the partial discharge soundprint signal database is used to store partial discharge soundprint feature information under different operating states and form a sample data set;

[0008] Based on the sample data set, using a gated recurrent neural network as the core to construct a first partial discharge soundprint diagnosis model, and at the same time combining a spatio-temporal attention mechanism to deeply extract and analyze the time-domain and spatial features of the soundprint signal to obtain an optimized second partial discharge soundprint diagnosis model;

[0009] Apply the trained second-stage voiceprint diagnosis model to the on-line monitoring of solid-insulated high-voltage systems, and process the real-time collected voiceprint signals. The processing process includes removing environmental noise and extracting characteristic parameters to obtain the extracted signal characteristics, and then comparing the extracted signal characteristics with the partial discharge signal characteristics in the database to determine whether they are consistent. If they are consistent, a partial discharge warning message is sent in real time. If they are not consistent, the comparison continues.

[0010] Preferably, use the voiceprint acquisition device to collect the ultrasonic voiceprint signals generated during the operation of the solid-insulated high-voltage system, including:

[0011] Collect the partial discharge signal S(t) of the operating solid-insulated high-voltage system through the voiceprint acquisition device. This signal is composed of the target partial discharge signal S p (t) and environmental noise S n (t), that is:

[0012] S(t) = S p (t) + S n (t)

[0013] Preferably, construct the partial discharge voiceprint signal database of the solid-insulated high-voltage system. The calculation formula of the partial discharge voiceprint signal database is as follows:

[0014]

[0015] In the formula, D is a set containing signals at multiple moments, S is the characteristic signal of the partial discharge signal, is the specific manifestation of the partial discharge signal in the time domain under different working conditions, and N is the number of partial discharge signal samples under different working conditions;

[0016] Preferably, perform processing through the blind source separation algorithm, and the processing process includes:

[0017] Based on the independent component analysis algorithm, solve the separation matrix W for the mixing matrix X of the signal S(t), which needs to satisfy:

[0018] S p (t) = W · X

[0019] In the formula, W is the separation matrix, X is the mixing matrix, and S p (t) is the value of the partial discharge signal p at time t;

[0020] Preferably, use the voiceprint acquisition device to collect the ultrasonic voiceprint signals generated during the operation of the solid-insulated high-voltage system to obtain the original voiceprint data containing partial discharge information, including:

[0021] Capture the first acoustic wave signal collected by the ultrasonic sensor arranged on the high-voltage busbar; capture the second acoustic wave signal collected by the high-sensitivity sensor on the weak area of the insulation structure; capture the third acoustic wave signal collected by the vibration sensor on the high-voltage box shell;

[0022] After the collected first acoustic wave signal, second acoustic wave signal and third acoustic wave signal are processed by analog-to-digital conversion, the original voiceprint data containing partial discharge information is obtained.

[0023] Preferably, the first partial discharge voiceprint diagnosis model is constructed with a gated recurrent neural network as the core, and at the same time, a spatio-temporal attention mechanism is combined to deeply extract and analyze the time-domain and spatial features of the voiceprint signal, and the optimized second partial discharge voiceprint diagnosis model is obtained, which includes:

[0024] The unit update formula of the gated recurrent neural network model is h t =z t The formula is:

[0025]

[0026] Among them, z t is the update gate, a value between 0 and 1, used to control the weight of the previous hidden state h t-1 and the currently calculated hidden state at the current moment, r t is the reset gate, also between 0 and 1, controlling the degree of influence of the previous hidden state h t-1 If r t is 1, the previous hidden state is completely passed; if it is 0, the previous hidden state is ignored, h t is the candidate hidden state at the current moment, x t is the input signal at the current moment;

[0027] The spatio-temporal attention mechanism calculates the attention weight α through the following formula i :

[0028]

[0029] Among them, h i is the time-domain feature, t i is the frequency-domain feature, v, W S , W t , b are all trainable parameters.

[0030] Preferably, the extracted signal features are compared with the partial discharge signal features in the database, which includes:

[0031] The second voiceprint diagnosis model completed through training performs similarity matching on the signal feature vector and the signal features in the database, and calculates the matching degree ρ:

[0032]

[0033] In the formula, F is the signal feature vector, and ρ is the matching degree. is the sum of the products of the corresponding position elements of the first signal and the second signal. is used to calculate the "size" of each signal, that is, their modulus length, representing the amplitude of the signal.

[0034] When ρ > ρ th it is determined as a partial discharge signal.

[0035] In a second aspect, the present application further provides a monitoring and protection strategy system for a vehicle-mounted solid insulation high-voltage system, based on the monitoring and protection strategy method for a vehicle-mounted solid insulation high-voltage system described in claim 1, including:

[0036] An acquisition module: used to use a voiceprint acquisition device to collect ultrasonic voiceprint signals generated during the operation of the solid insulation high-voltage system in real time, and obtain original voiceprint data containing partial discharge information;

[0037] A processing module: used to process the original voiceprint data through a blind source separation algorithm to obtain the partial discharge voiceprint signal of the solid insulation high-voltage system;

[0038] A construction module: used to construct a partial discharge voiceprint signal database of the solid insulation high-voltage system according to the obtained partial discharge voiceprint signal, where the partial discharge voiceprint signal database is used to store partial discharge voiceprint feature information in different operating states and form a sample data set;

[0039] An extraction and analysis module: used to construct a first partial discharge voiceprint diagnosis model based on the sample data set with a gated recurrent neural network as the core, and at the same time combine a spatio-temporal attention mechanism to deeply extract and analyze the time-domain and space-domain features of the voiceprint signal to obtain an optimized second partial discharge voiceprint diagnosis model;

[0040] A judgment module: used to apply the trained second partial discharge voiceprint diagnosis model to the online monitoring of the solid insulation high-voltage system, process the real-time collected voiceprint signal, where the processing process includes removing environmental noise and extracting characteristic parameters to obtain the extracted signal features, and then comparing the extracted signal features with the partial discharge signal features in the database to determine whether they are consistent. If they are consistent, a partial discharge warning message is sent in real time. If they are not consistent, the comparison continues.

[0041] In a third aspect, the present application further provides a monitoring and protection strategy device for a vehicle-mounted solid insulation high-voltage system, including:

[0042] A memory for storing a computer program;

[0043] A processor for implementing the steps of the monitoring and protection strategy method of the on-vehicle solid insulation high-voltage system when executing the computer program.

[0044] In a fourth aspect, the present application further provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned monitoring and protection strategy method based on the on-vehicle solid insulation high-voltage system are implemented.

[0045] The beneficial effects of the present invention are as follows:

[0046] By introducing the blind source separation algorithm, deep learning neural network and spatio-temporal attention mechanism, the present invention solves the problems of noise interference, incomplete signal feature extraction and slow diagnosis speed existing in the partial discharge diagnosis of traditional solid insulation high-voltage systems. It realizes high-precision and high-efficiency partial discharge sound pattern diagnosis, provides reliable support for the operation state monitoring of solid insulation high-voltage equipment, helps to extend the service life of the equipment, and improves the safety and stability of the power system.

[0047] Other features and advantages of the present invention will be described in the subsequent description, and part of them will become obvious from the description, or be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the structures specifically pointed out in the written description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1 It is a schematic flow chart of the monitoring and protection strategy method of the on-vehicle solid insulation high-voltage system described in the embodiments of the present invention;

[0050] Figure 2 It is a schematic structural diagram of the monitoring and protection strategy system of the on-vehicle solid insulation high-voltage system described in the embodiments of the present invention;

[0051] Figure 3 It is a schematic structural diagram of the monitoring and protection strategy device of the on-vehicle solid insulation high-voltage system described in the embodiments of the present invention.

[0052] In the figure: 701, acquisition module; 702, processing module; 703, construction module; 704, extraction and analysis module; 705, judgment module; 800, monitoring and protection strategy device for on-vehicle solid insulation high-voltage system; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Specific implementation mode

[0053] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein usually can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0054] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for differential description and cannot be construed as indicating or implying relative importance.

[0055] Embodiment 1:

[0056] This embodiment provides a method for monitoring and protecting the strategy of an on-vehicle solid insulation high-voltage system.

[0057] See Figure 1 , the figure shows that this method includes step S100, step S200, step S300, step S400 and step S500.

[0058] S100. Use a voiceprint acquisition device to collect ultrasonic voiceprint signals generated during the operation of the solid insulation high-voltage system in real time, and obtain original voiceprint data containing partial discharge information.

[0059] It should be noted that in this embodiment, the acquisition and separation of partial discharge signals of the solid high-voltage box of the EMU: During the operation of the solid high-voltage box of the EMU, due to its complex internal structure and high electric field environment, partial discharge phenomena are inevitable. To accurately collect partial discharge signals, ultrasonic sensor arrays and high-speed data acquisition devices (sampling rate ≥ 10MS / s) are installed inside and on the shell of the solid high-voltage box of the EMU in the present invention. The installation positions of the acquisition system are as follows:

[0060] Two ultrasonic sensors are arranged near the high-voltage busbar to capture acoustic signals caused by partial discharge; additional high-sensitivity sensors are installed in the weak area of the insulation structure (shield end) to enhance the monitoring coverage; vibration sensors are installed on the outer shell of the high-voltage box to capture the response waveform after the acoustic signal reaches the outer shell for auxiliary positioning. The collected signals are converted by analog-to-digital conversion and then enter the signal processing module. Due to the vibration, electromagnetic interference and external noise (track noise) in the operation environment of the EMU, the collected signals are usually the superposition of the partial discharge signal x and the noise.

[0061] In order to separate the pure partial discharge signal, the present invention adopts the blind source separation method, combines the prior knowledge of the specific noise spectrum of the EMU, and uses wavelet transform to denoise the signal. The processing steps include: wavelet decomposition, threshold processing, and signal reconstruction. After the above processing, the obtained signal not only retains the key characteristics of partial discharge but also eliminates the external noise interference, providing a reliable basis for subsequent modeling.

[0062] It can be understood that in this step, it includes S101:

[0063] Collect the partial discharge signal S(t) of the solid insulation high-voltage system in operation through the voiceprint acquisition device. This signal is composed of the target partial discharge signal S p (t) and the environmental noise S n (t), that is:

[0064] S(t) = S p (t) + S n (t)

[0065] This step also includes:

[0066] S102. Capture the first acoustic signal collected by the ultrasonic sensor arranged on the high-voltage busbar; capture the second acoustic signal collected by the high-sensitivity sensor on the weak area of the insulation structure; capture the third acoustic signal collected by the vibration sensor on the outer shell of the high-voltage box;

[0067] S103. After the collected first acoustic signal, second acoustic signal and third acoustic signal are processed by analog-to-digital conversion, the original voiceprint data containing partial discharge information is obtained.

[0068] It should be noted that in the actual operating environment, the collected partial discharge signals not only contain the target partial discharge signals but also are mixed with environmental noises. These noises may come from the vibrations of high-voltage equipment, electromagnetic interference, or other external factors. The voiceprint acquisition device can initially filter out some noises through the built-in noise reduction algorithm and self-developed DSP technology to ensure the quality of the collected signals. Additionally, in a high-voltage system, sensors arranged at different positions can capture different types of signals. The ultrasonic sensors on the high-voltage busbars are mainly used to monitor the partial discharge signals during high-voltage transmission; the highly sensitive sensors in the areas with weak insulation structures can capture subtle partial discharge signals to detect potential fault points in advance; the vibration sensors on the outer shell of the high-voltage box are used to monitor the mechanical vibrations during equipment operation to assist in judging the authenticity of the partial discharge signals. Among them, the collected acoustic signals are analog signals and need to be converted into digital signals through analog-to-digital conversion (ADC) for subsequent digital signal processing. During the analog-to-digital conversion process, it is necessary to ensure that the sampling frequency meets the Nyquist theorem to avoid signal distortion. Modern voiceprint acquisition devices usually adopt high-precision ADCs that can process high-frequency and wide dynamic range signals.

[0069] S200. Based on the original voiceprint data, through the processing of the blind source separation algorithm, the partial discharge voiceprint signal of the solid insulation high-voltage system is obtained.

[0070] It can be understood that in this step, the process of processing through the blind source separation algorithm includes:

[0071] Based on the independent component analysis algorithm, the separation matrix W is solved for the mixing matrix X of the signal S(t), where it is necessary to satisfy:

[0072] S p (t) = W·X

[0073] In the formula, W is the separation matrix, X is the mixing matrix, and S p (t) is the value of the partial discharge signal p at time t;

[0074] S300. According to the obtained partial discharge voiceprint signal, a partial discharge voiceprint signal database of the solid insulation high-voltage system is constructed, where the partial discharge voiceprint signal database is used to store the partial discharge voiceprint feature information under different operating states and form a sample data set.

[0075] In this embodiment, the construction of the voiceprint signal database of the solid high-voltage box of the EMU: A standardized partial discharge voiceprint signal database of the solid high-voltage box of the EMU is established, and data collection and collation are mainly carried out from the following aspects:

[0076] 1. Signal feature extraction: Extract the following key features of the partial discharge signal: time-domain features (signal amplitude, rise time, signal duration), frequency-domain features (peak frequency, spectral width, energy concentration), and time-frequency features (extract the time-frequency distribution diagram through the short-time Fourier transform (STFT) to analyze the coupling characteristics of the partial discharge signal in the frequency domain and time domain).

[0077] 2. Data grouping and annotation: Classify and annotate the signals according to different operating states (start-up, acceleration, cruise, braking) and different environments (high cold, high temperature, high humidity) of the EMU. The database contains two types of data: normal operating state and partial discharge abnormal state, and the specific abnormal positions (busbar, insulation support components) are annotated.

[0078] 3. Data storage: Store the above features and signal waveform data into the database in the format of JSON files for subsequent rapid retrieval and analysis.

[0079] It can be understood that in this step, a partial discharge soundprint signal database of the solid insulation high-voltage system is constructed, and the calculation formula of the partial discharge soundprint signal database is as follows:

[0080]

[0081] In the formula, D is a set containing signals at multiple moments, S is the characteristic signal of the partial discharge signal, is the specific manifestation of the partial discharge signal in the time domain under different working conditions, and N is the number of partial discharge signal samples under different working conditions;

[0082] Among them, in this step, the construction of the soundprint signal database D includes:

[0083] Extract the characteristic parameter vector F = [f 1 , f 2 , f 3 ,,,,, f k of the partial discharge soundprint signal, where f k includes time-domain features (such as signal amplitude, peak factor), frequency-domain features (such as main frequency, energy spectral density), and time-frequency domain joint features;

[0084] Furthermore, use the parameterization method to normalize the signal features:

[0085]

[0086] Among them, μ k is the feature mean, and σ k is the feature standard deviation.

[0087] Based on the sample data set, a first partial discharge acoustic fingerprint diagnosis model is constructed with a gated recurrent neural network as the core. At the same time, combined with the spatio-temporal attention mechanism, the time-domain and spatial features of the acoustic fingerprint signal are deeply extracted and analyzed to obtain an optimized second partial discharge acoustic fingerprint diagnosis model.

[0088] It should be noted that the gated recurrent neural network controls the flow of information by introducing gating mechanisms (such as reset gates and update gates), so as to effectively capture the time-series features in the acoustic fingerprint signal. When constructing the model, the core role of the GRNN is to process the time dynamics of the acoustic fingerprint signal and generate an embedding vector representing the partial discharge characteristics. The spatio-temporal attention mechanism is an advanced feature extraction technology that can capture the time-domain and spatial features of the acoustic fingerprint signal at the same time. The spatial attention mechanism can focus on the spatial distribution features of the signal, while the time attention mechanism focuses on the time continuity of the signal. Through this mechanism, the model can automatically identify and focus on the parts that contribute most to the recognition of partial discharge signals, ignoring unimportant features, thereby improving the accuracy and efficiency of diagnosis.

[0089] It can be understood that in this step, it includes:

[0090] The unit update formula of the gated recurrent neural network model h t = z t The formula is:

[0091]

[0092] Among them, z t is the update gate, a value between 0 and 1, used to control the weight of the previous hidden state h t-1 and the currently calculated hidden state at the current moment. r t is the reset gate, also between 0 and 1, controlling the degree of influence of the previous hidden state h t-1 . If r t is 1, the previous hidden state is completely passed; if it is 0, the previous hidden state is ignored. h t is the candidate hidden state at the current moment, x t is the input signal at the current moment;

[0093] The spatio-temporal attention mechanism calculates the attention weight α through the following formula i :

[0094]

[0095] Among them, h i is the time-domain feature, t i is the frequency-domain feature, v, W S , W t, both a and b are trainable parameters.

[0096] It should be noted that in this implementation, a gated recurrent unit (GRU) is used as the main body, combined with a spatiotemporal attention mechanism, to construct an intelligent diagnosis model, and the diagnosis ability of the model is improved through various optimization strategies. The specific implementation is as follows:

[0097] 1. Data input and feature engineering

[0098] To improve the model's representation ability of partial discharge signals, the data input into the model needs to undergo a series of feature engineering processes. First, the original signals are extracted from the constructed partial discharge soundprint signal database, and the short-time Fourier transform (STFT) is used to convert the time-domain signals into time-frequency diagrams, and the time-domain, frequency-domain, and time-frequency domain features of the signals are extracted. The time-domain features include statistical indicators such as signal amplitude, mean, variance, and skewness; the frequency-domain features include main frequency, spectral centroid, bandwidth, etc.; the time-frequency features are input in the form of two-dimensional time-frequency diagrams. Through this multi-dimensional feature fusion, the characteristics of partial discharge signals are fully retained, and the model's perception ability of different discharge modes is enhanced.

[0099] 2. Model structure design

[0100] The partial discharge diagnosis model takes the gated recurrent neural network as the core and combines the spatiotemporal attention mechanism to perform deep learning and feature optimization on complex signals:

[0101] Among them, the GRU network main body is an improved recurrent neural network (RNN). By introducing the update gate and reset gate mechanisms, the problem of gradient disappearance in traditional RNNs is solved. This model can model the time-series features of input signals and capture the short-term and long-term dependencies of partial discharge signals. In the present invention, the input of GRU is the multi-dimensional signal features after feature engineering, the network depth is 2-3 layers, and each layer contains 128-256 units.

[0102] Among them, for the spatiotemporal attention mechanism, to further improve the model's attention ability to the key features of partial discharge signals, the present invention introduces a spatiotemporal attention mechanism. The time attention mechanism captures the time key points (such as discharge mutation points) in the signal sequence by dynamically allocating weights; the spatial attention mechanism emphasizes the importance of feature dimensions, automatically allocates feature weights, and highlights the key frequency bands and amplitude change regions. After combining the spatiotemporal attention mechanism, the model can screen out the core features of partial discharge signals under the background of big data, effectively improving the diagnosis accuracy.

[0103] 3. Model training and verification

[0104] The constructed partial discharge diagnosis model needs to optimize the network parameters through the training process. The training process mainly includes the following steps:

[0105] 3.1. Data division: Extract the dataset from the partial discharge acoustic fingerprint signal database and divide it into a training set, a validation set, and a test set according to the ratio of 7:2:1 to ensure the fairness and effectiveness of the model training process.

[0106] 3.2. Training process: Use the AdamW optimizer with adaptive learning rate for gradient optimization, and select the weighted cross-entropy loss as the loss function to handle the imbalance problem of the number of data samples under different fault modes. During the model training process, introduce a dynamic learning rate adjustment strategy to reduce the learning rate when the model performance converges slowly, improving the convergence speed and stability.

[0107] 3.3. Regularization and overfitting prevention: To prevent the model from overfitting, add a Dropout layer after each GRU network layer to randomly discard some units to reduce the over-reliance on specific features; at the same time, adopt the early stopping strategy, and terminate the training when the performance on the validation set no longer improves.

[0108] 3.4. Model optimization and enhancement: To further improve the diagnostic performance of the model, the present invention adopts the following optimization strategies: First is transfer learning, that is, during the initial model training, use the partial discharge signal data of existing high-voltage power equipment for transfer learning to accelerate model convergence and improve diagnostic capabilities; second is data augmentation: Generate new partial discharge signal samples by adding noise, time shifting, time reversal, etc., expand the scale of training data, and enhance the model's adaptability to complex scenarios; then is ensemble learning: Introduce multiple deep learning models and GRUs to form an integrated diagnosis framework, and improve the robustness of the diagnostic results through multi-model voting.

[0109] 3.5. Real-time diagnosis and deployment: Finally, the trained partial discharge diagnosis model is deployed in the monitoring system of the solid high-voltage box of the EMU to collect and analyze partial discharge signals in real time. The model can automatically classify the input signals and output diagnostic results such as "normal", "slight discharge", "severe discharge", etc., and trigger corresponding alarm mechanisms according to the diagnostic results. Its real-time performance benefits from the efficient computing power of the GRU network and the feature screening advantages of the spatio-temporal attention mechanism, which can quickly process complex signal inputs and ensure the safe operation of the equipment.

[0110] Through the above implementation methods, the partial discharge diagnosis model constructed by the present invention shows significant advantages in terms of diagnostic accuracy, robustness, and real-time performance, providing a solid technical foundation for the intelligent monitoring of the solid insulation high-voltage system of the EMU.

[0111] S500. Apply the trained second-stage voiceprint diagnosis model to the on-line monitoring of the solid-insulated high-voltage system, process the real-time collected voiceprint signals, where the processing process includes removing environmental noise and extracting characteristic parameters, obtain the extracted signal characteristics, and then compare the extracted signal characteristics with the partial discharge signal characteristics in the database to determine whether they are consistent. If they are consistent, send out a partial discharge warning message in real time. If they are not consistent, continue the comparison.

[0112] It can be understood that in this step, the comparison of the extracted signal characteristics with the partial discharge signal characteristics in the database includes:

[0113] Perform similarity matching on the signal feature vector and the signal characteristics in the database through the trained second-stage voiceprint diagnosis model, and calculate the matching degree ρ:

[0114]

[0115] In the formula, F is the signal feature vector, and ρ is the matching degree. is the sum of the products of the corresponding elements of the first signal and the second signal. is to calculate the "size" of each signal, that is, their modulus length, representing the amplitude of the signal.

[0116] When ρ > ρ th it is judged as a partial discharge signal.

[0117] It should be noted that the processing of the real-time collected voiceprint signals includes:

[0118] Perform joint time-frequency domain analysis on the voiceprint signal by using short-time Fourier transform to obtain a two-dimensional spectrogram of time and frequency. The calculation formula is as follows:

[0119]

[0120] Among them,,,, ω(τ) is the window function, X(t,f) is the time-frequency spectrum, S(τ) is the original signal of the input signal at time τ, ω(τ - t) is the window function, used to limit the time window of the signal, that is, only focus on the time period near t when performing Fourier transform, e -j2πfτ dτ is the core term in the Fourier transform, representing the frequency component of the signal.

[0121] During operation, the solid insulation high-voltage system of the multiple unit train collects partial discharge signal data in real time through installed acoustic sensors and data acquisition modules. These signals are transmitted to the monitoring center via industrial Ethernet or wireless transmission modules and are preprocessed before entering the diagnostic model, including signal denoising, normalization, and feature enhancement, to ensure the quality and stability of the input signals, thus providing reliable data support for subsequent diagnosis. It should be noted that in the diagnostic process, the system first loads the acoustic fingerprint diagnostic model trained in the previous steps and initializes the model parameters, including the GRU network weights and configurations related to the spatio-temporal attention mechanism. The diagnostic threshold is also set during the initialization process, such as the discrimination criteria for normal signals, slight discharge signals, and severe discharge signals, to ensure that the model can output accurate diagnostic results according to the established classification rules.

[0122] Furthermore, the partial discharge signals that need to be collected in real time are analyzed through the model. This includes: First, the time series features of the signals are extracted by the GRU network, and the model can capture the dynamic changes of the signals and the time series patterns of partial discharge behaviors. Then, combined with the spatio-temporal attention mechanism, the model further focuses on the key time points and spectral features in the signals, analyzes important discharge information and abnormal behavior features. Through these steps, the diagnostic model generates discharge classification results, including status information such as "no discharge", "slight discharge", or "severe discharge", and feeds them back to the monitoring system interface in real time, providing clear status indications for operators.

[0123] When the diagnostic result shows abnormal discharge, the system automatically triggers a multi-level alarm mechanism. Visual alarms will highlight the abnormal equipment in red on the monitoring interface, accompanied by sound alarm prompts. In addition, the system will record the diagnostic results of the abnormal signals and the time-frequency feature data in the database to provide data support for subsequent abnormal analysis and fault tracing. If the discharge situation is relatively serious, the system will also send detailed alarm information to the maintenance personnel through the communication module, including the equipment location, abnormal signal characteristics, and recommended treatment plans, to ensure timely response and handling.

[0124] To improve the adaptability and diagnostic ability of the system, a dynamic learning mechanism is also designed for the system. During operation, abnormal data is automatically stored and labeled for expanding the acoustic fingerprint signal database. The system uses this new data to perform online training and optimization on the acoustic fingerprint diagnostic model, continuously updating the model parameters, thereby improving its diagnostic ability for complex scenarios and new discharge patterns. Through dynamic learning and optimization, the system can maintain the efficiency and accuracy of the diagnostic model, further enhancing the operation safety and intelligent level of the solid insulation high-voltage system.

[0125] It should be noted that in this step, when the diagnostic model detects a partial discharge signal, a partial discharge diagnostic report is automatically generated, including the intensity, frequency range, and duration of the partial discharge signal.

[0126] The method used in the present invention is applicable to different forms of solid insulation high-voltage systems, including vehicle-mounted high-voltage switchgear, power transformers and combined electrical equipment, and has wide applicability.

[0127] In summary, the present invention uses a voiceprint acquisition device to collect ultrasonic voiceprint signals in real time during the operation of a solid insulation high-voltage system, and obtains original voiceprint data containing partial discharge information; then, a blind source separation algorithm is used to process the original signal, and the target partial discharge voiceprint signal is separated from a variety of noise signals to obtain a clean solid insulation high-voltage system partial discharge voiceprint signal. Based on the above signal data, a partial discharge voiceprint signal database of a solid insulation high-voltage system is constructed to store the partial discharge voiceprint feature information under different operating conditions, forming a rich sample data set. On this basis, a partial discharge voiceprint diagnosis model is constructed using a gated recurrent neural network (GRNN) as the core, and the time domain and spatial features of the voiceprint signal are deeply extracted and analyzed in combination with a spatiotemporal attention mechanism, thereby significantly improving the model's ability to extract complex signal features and diagnostic accuracy. Furthermore, by training the model of partial discharge voiceprint data, the parameters of the diagnostic model are optimized to ensure that it has a high recognition ability and adaptability to partial discharge voiceprint signals in the actual operating environment. The trained diagnostic model is applied to the online monitoring of solid insulation high-voltage systems to process the real-time collected voiceprint signals, including the removal of environmental noise and the extraction of characteristic parameters. By comparing the extracted signal features with the partial discharge signal features in the database, when the diagnostic model identifies a partial discharge signal, a partial discharge alarm message is issued in real time.

[0128] Therefore, the method of the present invention realizes the accurate detection and diagnosis of partial discharge of solid insulation high-voltage system by constructing a complete voiceprint database and intelligent diagnosis model, which provides a reliable basis for the operation status monitoring and fault prediction of the high-voltage system, and reduces the difficulty and error of manual detection, and has broad application prospects.

[0129] Embodiment 2:

[0130] like Figure 2 As shown, this embodiment provides a monitoring and protection strategy system for a vehicle-mounted solid insulation high voltage system, see Figure 2 The system comprises:

[0131] Acquisition module 701: used to collect ultrasonic soundprint signals generated during the operation of the solid insulation high-voltage system in real time using a soundprint acquisition device to obtain original soundprint data containing partial discharge information;

[0132] Processing module 702: used to obtain the partial discharge voiceprint signal of the solid insulation high voltage system based on the original voiceprint data through blind source separation algorithm;

[0133] Building module 703: It is used to construct a partial discharge acoustic fingerprint signal database of the solid insulation high-voltage system according to the obtained partial discharge acoustic fingerprint signal. The partial discharge acoustic fingerprint signal database is used to store the partial discharge acoustic fingerprint feature information under different operating states and form a sample data set;

[0134] Extraction and analysis module 704: It is used to construct a first partial discharge acoustic fingerprint diagnosis model based on the sample data set with a gated recurrent neural network as the core, and at the same time combine a spatio-temporal attention mechanism to deeply extract and analyze the time-domain and spatial features of the acoustic fingerprint signal, and obtain an optimized second partial discharge acoustic fingerprint diagnosis model;

[0135] Judgment module 705: It is used to apply the trained second partial discharge acoustic fingerprint diagnosis model to the online monitoring of the solid insulation high-voltage system, process the real-time collected acoustic fingerprint signal. The processing process includes removing environmental noise and extracting characteristic parameters to obtain the extracted signal features, and then comparing the extracted signal features with the partial discharge signal features in the database to judge whether they are consistent. If they are consistent, a partial discharge warning message will be sent in real time. If they are not consistent, the comparison will continue.

[0136] It should be noted that for the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0137] Embodiment 3:

[0138] Corresponding to the above method embodiment, in this embodiment, a monitoring and protection strategy device for a vehicle-mounted solid insulation high-voltage system is also provided. The monitoring and protection strategy device for a vehicle-mounted solid insulation high-voltage system described below can be mutually referred to the monitoring and protection strategy method for a vehicle-mounted solid insulation high-voltage system described above.

[0139] Figure 3 It is a block diagram of a monitoring and protection strategy device 800 for a vehicle-mounted solid insulation high-voltage system shown according to an exemplary embodiment. As Figure 3 shown, the monitoring and protection strategy device 800 for the vehicle-mounted solid insulation high-voltage system includes: a processor 801 and a memory 802. The monitoring and protection strategy device 800 for the vehicle-mounted solid insulation high-voltage system also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0140] Among them, the processor 801 is used to control the overall operation of the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system to complete all or part of the steps in the above-mentioned monitoring and protection strategy method of the vehicle-mounted solid insulation high-voltage system. The memory 802 is used to store various types of data to support the operation of the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system. These data may include, for example, instructions for any application or method operating on the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, and so on. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, magnetic disk or optical disc. The multimedia component 803 may include a screen and an audio component. The screen can be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse or buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them. Accordingly, the communication component 805 may include: a Wi-Fi module, a Bluetooth module or an NFC module.

[0141] In an exemplary embodiment, the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system can be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components, and is used to execute the above-mentioned monitoring and protection strategy method of the vehicle-mounted solid insulation high-voltage system.

[0142] In another exemplary embodiment, a computer-readable storage medium including program instructions is further provided. When the program instructions are executed by a processor, the steps of the above-mentioned monitoring and protection strategy method of the vehicle-mounted solid insulation high-voltage system are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above program instructions can be executed by the processor 801 of the monitoring and protection strategy device 800 of the vehicle-mounted solid insulation high-voltage system to complete the above-mentioned monitoring and protection strategy method of the vehicle-mounted solid insulation high-voltage system.

[0143] Embodiment 4:

[0144] Corresponding to the above method embodiment, in this embodiment, a readable storage medium is further provided. A readable storage medium described below can be correspondingly referred to with a monitoring and protection strategy method of a vehicle-mounted solid insulation high-voltage system described above.

[0145] A computer program is stored on the readable storage medium. When the computer program is executed by a processor, the steps of the monitoring and protection strategy method of the vehicle-mounted solid insulation high-voltage system in the above method embodiment are implemented.

[0146] The readable storage medium can specifically be various readable storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc that can store program codes.

[0147] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0148] As described above, this is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or replacements, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system, characterized in that: include: The ultrasonic soundprint signal generated during the operation of the solid insulation high-voltage system is collected in real time by using a soundprint collection device to obtain the original soundprint data containing partial discharge information; Based on the original voiceprint data, the partial discharge voiceprint signal of the solid insulation high-voltage system is obtained through blind source separation algorithm processing; According to the obtained partial discharge soundprint signal, a partial discharge soundprint signal database of the solid insulation high-voltage system is constructed, wherein the partial discharge soundprint signal database is used to store the partial discharge soundprint feature information under different operating conditions and form a sample data set; Based on the sample data set, the first partial release voiceprint diagnosis model is constructed with the gated recurrent neural network as the core. At the same time, the time domain and spatial features of the voiceprint signal are deeply extracted and analyzed in combination with the spatiotemporal attention mechanism to obtain the optimized second partial release voiceprint diagnosis model. The trained second partial discharge soundprint diagnostic model is applied to the online monitoring of solid insulation high-voltage system, and the real-time collected soundprint signal is processed. The processing process includes the removal of environmental noise and the extraction of characteristic parameters to obtain the extracted signal characteristics. The extracted signal characteristics are then compared with the partial discharge signal characteristics in the database to determine whether they are consistent. If they are consistent, the partial discharge alarm information is issued in real time. If they are inconsistent, the comparison continues.

2. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1 is characterized in that: The use of the voiceprint collection device to collect ultrasonic voiceprint signals generated during the operation of the solid insulation high-voltage system in real time includes: The partial discharge signal S(t) of the solid insulation high voltage system in operation is collected by the voiceprint collection device. The signal is composed of the target partial discharge signal S p (t) and environmental noise S n (t) composed of: S(t)=S p (t)+S n (t) 3. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The partial discharge soundprint signal database of the solid insulation high voltage system is constructed, wherein the calculation formula of the partial discharge soundprint signal database is as follows: Where D is a set of multiple time signals, S is the characteristic signal of the partial discharge signal, is the specific performance of the partial discharge signal in the time domain under different working conditions, and N is the number of partial discharge signal samples under different working conditions.

4. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The blind source separation algorithm is used for processing, wherein the processing process includes: Based on the independent component analysis algorithm, the separation matrix W is solved for the mixing matrix X of the signal S(t), which needs to satisfy: S p (t)=W·X Where W is the separation matrix, X is the mixing matrix, S p (t) represents the value of the partial discharge signal p at time t.

5. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The ultrasonic soundprint signal generated during the operation of the solid insulation high-voltage system is collected in real time by using the soundprint collection device to obtain the original soundprint data containing partial discharge information, including: Capture the first sound wave signal collected by the ultrasonic sensor arranged on the high-voltage busbar; capture the second sound wave signal collected by the high-sensitivity sensor on the weak area of ​​the insulation structure; capture the third sound wave signal collected by the vibration sensor on the high-voltage box shell; The collected first sound wave signal, second sound wave signal and third sound wave signal are processed by analog-to-digital conversion to obtain original voiceprint data containing partial discharge information.

6. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The gated recurrent neural network is used as the core to construct the first partial discharge voiceprint diagnosis model, and the spatiotemporal attention mechanism is combined to deeply extract and analyze the time domain and spatial features of the voiceprint signal to obtain the optimized second partial discharge voiceprint diagnosis model, which includes: Unit update formula of gated recurrent neural network model t =z t The formula is: Among them, z t is the update gate, which is a value between 0 and 1 and is used to control the hidden state h of the previous moment t-1 and the currently computed hidden state The weight at the current moment, r t is the reset gate, also between 0 and 1, controlling the hidden state h of the previous moment t-1 If r t If it is 1, the hidden state of the previous moment is fully transmitted; if it is 0, the hidden state of the previous moment is ignored, h t is the candidate hidden state at the current moment, x t is the input signal at the current moment; The spatiotemporal attention mechanism calculates the attention weight α by the following formula i : Among them, h i is the time domain feature, t i is the frequency domain feature, v,W S ,W t , b are both trainable parameters.

7. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The extracted signal features are compared with the partial discharge signal features in the database, including: The trained second-partial release voiceprint diagnosis model performs similarity matching between the signal feature vector and the signal feature in the database, and calculates the matching degree ρ: In the formula, F is the signal feature vector, ρ is the matching degree, is the sum of the products of the first signal and the elements at corresponding positions in the second signal, To calculate the modulus of each signal, it represents the amplitude of the signal; When ρ>ρ th When , it is judged as a partial discharge signal.

8. The monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: The processing of the voiceprint signal collected in real time includes: Short-time Fourier transform is used to perform a joint analysis of the voiceprint signal in the time and frequency domains to obtain a two-dimensional spectrogram of time and frequency. The calculation formula is as follows: Where ω(τ) is the windowing function, X(t,f) is the time spectrum, S(τ) is the original input signal at time τ, and ω(τ-t) is the window function, which is used to limit the time window of the signal, that is, to focus on the time period near t when performing Fourier transform. -j2πfτ dτ is the core term in Rieger transform, which represents the frequency component of the signal.

9. A monitoring and protection strategy system for a vehicle-mounted solid insulation high voltage system, based on the monitoring and protection strategy method for a vehicle-mounted solid insulation high voltage system according to claim 1, characterized in that: include: Acquisition module: used to collect ultrasonic soundprint signals generated during the operation of the solid insulation high-voltage system in real time using a soundprint acquisition device to obtain original soundprint data containing partial discharge information; Processing module: used to obtain the partial discharge voiceprint signal of the solid insulation high-voltage system based on the original voiceprint data through blind source separation algorithm; Construction module: used to construct a partial discharge soundprint signal database of a solid insulation high-voltage system according to the obtained partial discharge soundprint signal, wherein the partial discharge soundprint signal database is used to store partial discharge soundprint feature information under different operating conditions and form a sample data set; Extraction and analysis module: It is used to build the first partial release voiceprint diagnosis model based on the sample data set, using the gated recurrent neural network as the core, and at the same time, combined with the spatiotemporal attention mechanism, deeply extract and analyze the time domain and spatial features of the voiceprint signal to obtain the optimized second partial release voiceprint diagnosis model; Judgment module: used to apply the trained second partial discharge soundprint diagnostic model to the online monitoring of solid insulation high-voltage system, and process the real-time collected soundprint signal, wherein the processing process includes the removal of environmental noise and the extraction of characteristic parameters, and obtains the extracted signal characteristics, and then compares the extracted signal characteristics with the partial discharge signal characteristics in the database to determine whether they are consistent. If they are consistent, the partial discharge alarm information is issued in real time. If they are inconsistent, the comparison continues.

10. A monitoring and protection strategy device for a vehicle-mounted solid insulation high voltage system, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the monitoring and protection strategy method for the on-board solid insulation high-voltage system as claimed in any one of claims 1 to 8 when executing the computer program.

Images