Discharge monitoring system and method based on dynamic frequency band selection and network identification

Abstract

The invention discloses a discharge monitoring system and method based on dynamic frequency band selection and network identification, and the system comprises a signal collection and processing module, a dynamic frequency band optimization module, and an intelligent identification and analysis module. The harmonic interference of communication frequency bands and industrial equipment is actively avoided, the background purity of partial discharge signals is improved, the inherent sensitivity defect of fixed frequency band monitoring is overcome through a millisecond-level frequency band seamless switching mechanism in a strong electromagnetic coupling scene of a transformer substation, the weak discharge signal capturing capacity is leaky, the leak detection phenomenon is inhibited, and the reliability of the system is improved. A reliable data basis is provided for evaluation of the insulation state of power equipment, rapid immune response and dynamic optimization capability are formed for novel electromagnetic interference, the system autonomously perfects a noise knowledge base and an identification boundary in long-term operation, and it is ensured that the monitoring performance is kept stable and reliable in the whole life cycle.

Description

Technical Field

[0001] The present invention relates to the technical field of on-line monitoring of power equipment, and particularly to a discharge monitoring system and method based on dynamic frequency band selection and network identification. Background Art

[0002] A power equipment monitoring device is a device used to monitor, detect and manage the state of power system equipment. It plays an important role in the operation and maintenance of the power system, helps to improve the reliability, safety and economy of the power system. The on-line monitoring device of power equipment can monitor the operation state of power equipment in real time, timely detect possible fault signs of the equipment, and issue an alarm in advance to avoid downtime and losses caused by equipment failures.

[0003] However, there are some problems in the prior art: 1. The limitation of fixed frequency band monitoring; 2. Poor adaptability to noise interference; 3. The prior art has not solved the collaborative optimization problem of dynamic frequency band selection and real-time signal validity verification.

[0004] Therefore, a discharge monitoring system and method based on dynamic frequency band selection and network identification are proposed to solve the above problems. Summary of the Invention

[0005] The main object of the present invention is to provide a discharge monitoring system and method based on dynamic frequency band selection and network identification to solve the problems raised in the above background.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a discharge monitoring method based on dynamic frequency band selection and network identification, the method comprising the following steps: Step S1: Collect the original electromagnetic signals of the power equipment through a UHF sensor array to generate an initial UHF signal data set; Step S2: Input the initial UHF signal data set into a multi-band filtering module for parallel frequency band processing to generate a low-frequency bandpass signal, a high-frequency bandpass signal, a full-frequency direct communication signal and a selective dual-band combination signal; Step S3: Respectively perform time-domain feature extraction, frequency-domain feature extraction and statistical feature extraction on the frequency band signals output in step S2 to generate a multi-dimensional feature vector set; Step S4: Input the multi-dimensional feature vector set into a pre-trained neural network classification model for signal validity verification, output the partial discharge signal probability value and the noise probability value, and generate signal classification identification data; Step S5: Calculate the real-time signal-to-noise ratio SNR of each frequency band based on the signal classification identification data, and generate a frequency band performance evaluation matrix in combination with a preset pulse density threshold; Step S6: Execute a dynamic frequency band switching decision according to the frequency band performance evaluation matrix; Step S7: Adjust the frequency band polling period according to the switching mode parameters configured by the user, generate an optimized partial discharge monitoring result, and output an alarm instruction; Preferably, the step S2 includes: Step S21: Configure four parallel filtering links through a programmable filter module: Link 1: Generate a low-frequency bandpass signal using a 550 MHz bandpass filter; Link 2: Generate a high-frequency bandpass signal using a 1300 MHz bandpass filter; Link 3: Generate a full-frequency direct-through signal using a 900 MHz all-pass filter; Link 4: Dynamically select two non-overlapping frequency bands for combined filtering to generate a dual-band combined signal; Step S22: Control the signal output timing of each filtering link through a multiplexer.

[0007] Preferably, the operation of frequency domain feature extraction in the step S3 is specifically: Step S31: Perform discrete wavelet transform on each frequency band signal and decompose it into 8 sub-bands; Step S32: Calculate the energy proportion of each sub-band: ; Where, is the energy proportion of the k-th sub-band, is the wavelet coefficient of the k-th sub-band, is the sampling point index, N is the number of sampling points, is the sub-band index; Step S33: Use the energy proportion distribution vector as the frequency domain feature.

[0008] Preferably, the construction of the neural network classification model in the step S4 includes: Step S41: The model structure adopts a CNN-LSTM hybrid architecture: Input layer: Receive a feature vector with a dimension of 12; Convolution layer: 3 layers of 1D-CNN, with a convolution kernel size of 3×1; LSTM layer: The number of bidirectional LSTM units is 64; Output layer: Output the partial discharge / noise probability through the Softmax activation function; Step S42: The training data includes: Positive samples: Pulse waveforms of corona discharge and floating discharge; Negative samples: Mobile phone signals, arc noise, and switch operation interference.

[0009] Preferably, the logic of dynamic frequency band switching decision in the step S6 is: Step S61: Initialize the polling period T and activate the signal acquisition of each frequency band in sequence; Step S62: Calculate the comprehensive performance score of each frequency band: ; Among them, is the comprehensive performance score of the i-th frequency band, is the current signal-to-noise ratio, is the switching time consumption, is the weight coefficient; Step S63: Select the frequency band with the highest

[0010] As an option, the switching mode parameter configuration in step S7 includes: Emergency mode: Polling period T = 1s, SNR drop threshold = 15%; Steady state mode: Polling period T = 15min, SNR drop threshold = 25%.

[0011] As an option, the method further includes an online learning mechanism: Step S71: When unrecognized signal types continuously appear, trigger the data enhancement module to generate synthetic samples; Step S72: Update the neural network classification model using the incremental learning algorithm: ; Among them, are the updated model parameters, are the model parameters before update, is the learning rate, is the gradient of the loss function L with respect to the parameter θ, is the predicted output of the model for the new sample , is the true label of the new sample.

[0012] As an option, the calculation formula for the signal-to-noise ratio SNR in step S5 is: ; Among them is the average power of the partial discharge pulse, is the noise power in the section without pulses.

[0013] As an option, the system includes: Signal acquisition and processing module, dynamic frequency band optimization module, intelligent identification and analysis module; The signal acquisition and processing module includes: Ultra-high frequency sensor array: Arranged at the pot-type insulator of the GIS equipment to collect 900 MHz electromagnetic signals; Signal conditioning unit: includes a preamplifier and a 24-bit ADC converter; The dynamic frequency band optimization module includes: Programmable multi-bandpass filter bank: hardware implementation of four parallel filtering links; Frequency band switching controller: real-time SNR evaluation and switching logic circuit based on FPGA; The intelligent recognition and analysis module includes: Feature extraction engine: performs time-domain, frequency-domain, and statistical feature calculations; Neural network processor: deploys a CNN-LSTM classification model; Online learning unit: supports incremental model updates.

[0014] Preferably, the programmable multi-bandpass filter bank is designed with a microstrip line structure and includes: Low-frequency path: center frequency 600 MHz, bandwidth ±200 MHz; High-frequency path: center frequency 1250 MHz, bandwidth ±250 MHz; Dual-frequency combination path: switches the frequency band combination method through PIN diodes.

[0015] The present invention has the following beneficial effects: 1. In the present invention, by applying a dynamic frequency band selection mechanism, when performing real-time monitoring of the discharge phenomenon of electrical equipment, by intelligently evaluating the signal-to-noise ratio and environmental interference level of different monitoring frequency bands, and setting an optimal monitoring frequency band selection strategy for specific discharge types, the clarity and pertinence of the acquisition process of different discharge signals are ensured. At the same time, the discharge signal is captured and analyzed in real time within the selected frequency band, and it can be judged in real time whether the selected frequency band can still effectively capture the target discharge characteristics, avoiding signal distortion or omission problems caused by interference in the fixed frequency band, ensuring the accuracy and integrity of the discharge signal acquisition, and reducing the monitoring blind area and false alarm risk. 2. In the present invention, by integrating the network recognition algorithm end, when performing discharge signal feature extraction and pattern discrimination, by calculating the discrimination degree between the discharge pulse waveform and the background noise and the aggregation degree in the feature space, the pattern category to which the discharge signal belongs is recognized in real time, so that the system can reduce the situation of misjudging noise or non-target signals as target discharges. And when it is found during the monitoring process that the signal characteristics deviate significantly from the preset pattern, the recognition result can be assisted or corrected in real time through the preset discharge feature map library, so that when the discharge pattern discrimination is abnormal, it can be quickly prompted or corrected, ensuring the accuracy and reliability of the discharge type recognition. 3. In the present invention, by constructing an intelligent alarm grading terminal, when comprehensively evaluating and warning the status of electrical equipment, after identifying an effective discharge signal, through multi-dimensional analysis and hierarchical evaluation of the intensity, frequency, development trend, and pattern evolution of the discharge signal, the deviation value of the hazard level of the current discharge activity is quantified in real time. According to the deviation values of the hazard levels in different dimensions, the health risks of the equipment insulation status are comprehensively evaluated and hierarchically warned in real time, enabling the system to achieve differentiated and stepped alarm management for different discharge types and severities, reducing the situation of a large number of invalid alarms or serious state omissions caused by the triggering of a single threshold by the monitoring system, and further improving the effectiveness of discharge monitoring, the accuracy of operation and maintenance guidance, and the overall monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of a discharge monitoring system and method based on dynamic frequency band selection and network identification according to the present invention; Figure 2 is a system flowchart of a discharge monitoring system and method based on dynamic frequency band selection and network identification according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1 - Figure 2 , the discharge monitoring method based on dynamic frequency band selection and network identification includes the following steps: Step S1: Collect the original electromagnetic signals of the power equipment through the UHF sensor array to generate an initial UHF signal data set; Step S2: Input the initial UHF signal data set into the multi-band filtering module for parallel frequency band processing to generate a low-frequency bandpass signal, a high-frequency bandpass signal, a full-frequency direct communication signal, and a selective dual-frequency combined signal; Step S3: Respectively perform time-domain feature extraction, frequency-domain feature extraction, and statistical feature extraction on the frequency band signals output in Step S2 to generate a multi-dimensional feature vector set; Step S4: Input the multi-dimensional feature vector set into a pre-trained neural network classification model for signal validity verification, output the partial discharge signal probability value and the noise probability value, and generate signal classification identification data; Step S5: Calculate the real-time signal-to-noise ratio SNR of each frequency band based on the signal classification identification data, and generate a frequency band performance evaluation matrix in combination with a preset pulse density threshold; Step S6, execute dynamic frequency band switching decision according to the frequency band performance evaluation matrix; Step S7, adjust the frequency band polling period according to the switching mode parameters configured by the user, generate an optimized partial discharge monitoring result and output an alarm instruction; Step S2 includes: Step S21, configure four parallel filtering links through the programmable filter module: Link 1: Generate a low-frequency band-pass signal using a 550 MHz band-pass filter; Link 2: Generate a high-frequency band-pass signal using a 1300 MHz band-pass filter; Link 3: Generate a full-frequency direct-through signal using a 900 MHz all-pass filter; Link 4: Dynamically select two non-overlapping frequency bands for combined filtering to generate a dual-frequency band combined signal; Step S22, control the signal output timing of each filtering link through the multiplexer; The operation of frequency domain feature extraction in Step S3 is specifically: Step S31, perform discrete wavelet transform on each frequency band signal and decompose it into 8 sub-frequency bands; Step S32, calculate the energy proportion of each sub-frequency band: ; where, is the energy proportion of the k-th sub-frequency band, is the wavelet coefficient of the k-th sub-frequency band, is the sampling point index, N is the number of sampling points, is the sub-frequency band index; Step S33, use the energy proportion distribution vector as the frequency domain feature; The construction of the neural network classification model in Step S4 includes: Step S41, the model structure adopts a CNN-LSTM hybrid architecture: Input layer: Receive a feature vector with a dimension of 12; Convolution layer: 3 layers of 1D-CNN, with a convolution kernel size of 3×1; LSTM layer: 64 bidirectional LSTM units; Output layer: Output the partial discharge / noise probability using the Softmax activation function; Step S42, the training data includes: Positive samples: Pulse waveforms of corona discharge and floating discharge; Negative samples: Mobile phone signals, arc noise, switch operation interference; The logic of dynamic frequency band switching decision in Step S6 is: Step S61, initialize the polling period T, and activate the signal acquisition of each frequency band in sequence; Step S62: Calculate the comprehensive performance scores for each frequency band: ; Among them, is the comprehensive performance score of the i-th frequency band, is the current signal-to-noise ratio, is the switching time consumption, is the weight coefficient; Step S63: Select the frequency band with the highest as the main monitoring channel; The switching mode parameter configuration in Step S7 includes: The system includes: Emergency mode: polling period T = 1s, SNR drop threshold = 15%; Steady state mode: polling period T = 15 min, SNR drop threshold = 25%; Step S71: When continuously unrecognized signal types appear, trigger the data enhancement module to generate synthetic samples; ; Among them, is the updated model parameter, is the model parameter before update, is the learning rate, is the gradient of the loss function L with respect to the parameter θ, is the predicted output of the model for the new sample and is the true label of the new sample; The calculation formula for the signal-to-noise ratio SNR in Step S5 is: ; Among them is the average power of the partial discharge pulse, is the noise power in the pulse-free section; The system includes: Signal acquisition and processing module, dynamic frequency band optimization module, intelligent identification and analysis module; The signal acquisition and processing module includes: Ultra-high frequency sensor array: Arranged at the pot-type insulator of the GIS device to collect 900 MHz electromagnetic signals; Signal conditioning unit: Comprising a preamplifier and a 24-bit ADC converter; The dynamic frequency band optimization module includes: Programmable multi-bandpass filter bank: Hardware implementation of four parallel filtering links; Frequency band switching controller: Real-time SNR evaluation and switching logic circuit based on FPGA; The intelligent recognition and analysis module includes: Feature extraction engine: performing time-domain, frequency-domain, and statistical feature calculations; Neural network processor: deploying a CNN-LSTM classification model; Online learning unit: supporting incremental model updates; The programmable multi-bandpass filter bank is designed with a microstrip line structure and includes: Low-frequency path: with a center frequency of 600 MHz and a bandwidth of ±200 MHz; High-frequency path: with a center frequency of 1250 MHz and a bandwidth of ±250 MHz; Dual-frequency combination path: switching the frequency band combination mode through PIN diodes.

[0019] Example 1: Insulation fault monitoring of GIS equipment 1. Signal acquisition and preprocessing Collect 900 MHz raw electromagnetic signals through a UHF sensor array annularly arranged on the basin insulator. After conditioning by a 40 dB gain low-noise amplifier, digital conversion is completed by a 500 MS / s sampling rate ADC module. 2. Dynamic frequency band optimization Initialize four filter channels: low-frequency bandpass, high-frequency bandpass, full-frequency direct-through, and dual-frequency combination; Analyze the ambient noise spectrum in real time. When 1.8 GHz communication interference is detected, automatically close the high-frequency channel; Calculate the signal-to-noise ratio of the remaining channels and lock the low-frequency channel as the main monitoring path.

[0020] 3. Feature fusion and recognition Extract the time-domain and frequency-domain features of the low-frequency channel signal; Input a 12-dimensional feature vector into the pre-trained CNN-LSTM model; If the probability of floating discharge output by the model is greater than the threshold, generate an insulation defect warning.

[0021] 4. Online learning trigger When unrecognized pulses continuously appear at the same location: Collect new samples and add time-shift jitter; Start incremental training to update the weights of the fully connected layer; Hot replace the classification model after the accuracy of the validation set reaches the standard.

[0022] Example 2: Transformer monitoring in plateau substations 1. Anti-interference initialization Deploy low-temperature-resistant sensors at a site with an altitude of 3000 meters. Automatically detect the corona noise background at startup and establish a baseline noise library.

[0023] 2. Response to Harsh Working Conditions Blizzard weather triggers the emergency mode, and the frequency band polling period is shortened to the millisecond level; The dual-band combined channel tracks the discharge pulse in real time; Filter the interference of ice drop through phase distribution statistics.

[0024] 3. Multi-dimensional Diagnostic Decision-making Fuse the oil chromatogram data and the discharge pulse density; Identify the development trajectory of the inter-turn discharge of the winding; Output the evaluation report of the insulation aging level.

[0025] Embodiment 3: Monitoring of Offshore Wind Power Converters 1. Adaptation to Salt Spray Environment Collect the characteristics of sea breeze salt spray corona in the initialization stage; Dynamically shield the high-interference frequency band of 1.3 GHz; Enable the full-frequency through mode to capture wide-spectrum discharges.

[0026] 2. Mechanical Vibration Decoupling Install vibration accelerometers for synchronous acquisition; Freeze the pulse count when the vibration frequency is greater than 100 Hz; 3. Life Prediction Linkage Correlate the operating duration of the converter with the trend of discharge intensity; Establish a remaining life model for the IGBT module; Warn the replacement window of the power device 8 weeks in advance.

[0027] Embodiment 4: Partial Discharge Monitoring of Large Hydropower Generators 1. Anti-interference Deployment in Strong Electromagnetic Environment Install high-temperature resistant fiber optic composite sensors in the stator slots of the generator. Use fluororubber seals to ensure stable insulation performance in a 95% humidity environment. The sensor array has 24 measurement points evenly distributed along the circumferential direction. Synchronously collect temperature-discharge dual parameters through fiber Bragg gratings. Perform reference calibration when the system starts, collect the harmonic interference characteristics of the excitation system under no-load conditions, and establish a background noise model including the rotor rotation modulation effect.

[0028] 2. Wide-frequency Domain Dynamic Sensing and Optimization Real-time scan the 1.5 GHz spectrum. When detecting the interference of the excitation carbon brush spark: Activate the high-frequency monitoring channel Inject an in-phase noise cancellation signal: ; Wherein, is the generated noise cancellation signal, A is the interference amplitude, is the center time point of the interference signal, is the Gaussian envelope width parameter, is the interference center frequency; Adopt wavelet packet reconstruction technology to enhance the effective frequency band: Perform 8-layer decomposition on the original signal; Select the energy-focused sub-band to reconstruct the signal; The signal-to-noise ratio is increased by more than 22 dB.

[0029] 3. Decoupling of multi-source discharge characteristics Establish a discharge pulse fingerprint library: Corona discharge of stator bars: The pulse rise time is greater than 15 ns, and the phase is dispersed; Slot wedge loosening discharge: The pulse cluster period is 30 ms, and the phase is concentrated; Creepage of cooling water pipes: accompanied by a temperature gradient greater than 3 °C / cm; Develop a spatio-temporal joint analysis algorithm: Spatial domain: Solve the discharge source coordinates based on the measurement point position matrix; ; Among them, is the spatial coordinate of the discharge source to be solved, is the coordinate of the i-th sensor, is the electromagnetic wave propagation speed, is the time delay of the discharge pulse reaching the i-th sensor; Time domain: Extract the phase relationship between the discharge pulse and the power frequency voltage; When the slot discharge characteristics are detected: Associate the vibration monitoring data; Output the slot wedge tightening operation instruction.

[0030] 4. Special response to the hydraulic environment High humidity environment: Start the microwave drying module to maintain the dew point of the sensor cavity less than -15 °C; Adopt the dielectric constant compensation algorithm to correct the pulse amplitude: ; Water quality deterioration during the flood season: Monitor the change of the conductivity of the cooling water; When surface discharge is detected: If the conductivity of the cooling water is greater than 10 μS / cm, it is determined that the insulator surface is contaminated and discharged; If the conductivity is normal but the discharge amount suddenly increases, a warning is issued for the stator bar water seepage; Low temperature and freezing conditions: The heating module maintains the sensor temperature greater than 5 °C; The pulse width discrimination technique is adopted to eliminate the interference of ice crystal rupture.

[0031] Example 5: Monitoring of valve towers in UHV converter stations 1. Deployment of multi-layer electromagnetic shielding In the ±800 kV valve hall, a UHF sensor is encapsulated with a carbon fiber composite shielding cover, and the signal line is led out through a waveguide penetrator. The sensor array is distributed according to the valve tower levels, and three groups of orthogonally polarized antennas are arranged at each level to cover the omnidirectional radiation field of the thyristor components. During startup, baseline calibration is performed, background noise without working conditions is collected, and an interference feature library including the harmonics of the cooling fan and the spikes of the thyristor switch is established.

[0032] 2. Dynamic spectrum sensing and avoidance The 1.8 GHz spectrum is scanned in real time. When the communication frequency point of the DC protection system is detected; The dual-band combination mode is activated; An adaptive notch filter is used to deeply suppress the 2.4 GHz ± 50 MHz frequency band; The signal-to-noise ratio gain within the effective monitoring bandwidth is calculated to be greater than 18 dB.

[0033] 3. Multi-physical field fusion diagnosis The cooling water flow rate and bus temperature data of the valve tower are synchronously collected; When abnormal discharge pulses are detected, a fault matrix is established by correlating the cooling parameters: If the flow rate decreases and the discharge phase is concentrated, it is determined that the waterway is blocked, resulting in local overheating and discharge; If the temperature is normal but the pulse repetition rate suddenly increases, a warning is issued for the insulation deterioration of the thyristor trigger unit; The discharge source is mapped to the specific level coordinates through a three-dimensional positioning algorithm.

[0034] 4. Emergency mechanism for extreme working conditions When the lightning strike causes the grounding grid potential to rise, the fiber optic transmission channel is automatically switched; The emergency monitoring mode is triggered: The polling period is compressed to 500 ms; Full-band scanning is enabled to capture transient discharges; The neural network model is switched to a lightweight version; The electromagnetic transient process and the discharge correlation characteristics are recorded, and an insulation strength attenuation curve is generated.

[0035] Example 6: Monitoring of power systems on offshore platforms 1. Adaptation to high salt fog environment Helium-sealed sensors are used, and a polytetrafluoroethylene anti-corrosion layer is sprayed on the surface; The characteristics of sea fog corona noise are collected during the initialization stage; The high-interference frequency bands are dynamically shielded, and the optimal monitoring window is locked.

[0036] 2. Mechanical vibration decoupling analysis Install a three-axis accelerometer to synchronously collect the vibration spectrum of the platform; Develop a vibration-discharge coupling separation algorithm: ; Among them, is the spectrum of the separated discharge signal, is the spectrum of the original mixed signal, is the spectrum of the vibration signal, is the transfer function, is the coupling coefficient; During the typhoon passing through, the effective pulse capture rate remains greater than 92%.

[0037] 3. Multi-energy system collaborative monitoring Associate the operating states of gas turbine generator sets, photovoltaic inverters, and energy storage converters; When abnormal discharge is detected: If the discharge occurs on the bus side of the gas turbine and is accompanied by an increase in CO concentration, warn of the pyrolysis of insulating materials; If the discharge is synchronized with the photovoltaic inverter switching frequency, diagnose the failure of the DC bus support capacitor; Establish a device health score model to drive preventive maintenance decisions.

[0038] Example 7: Monitoring of underground substations in the urban core area 1. Ultra-dense interference suppression Construct a panoramic electromagnetic environment map: 5G base station downlink frequency band Subway vehicle-ground communication frequency band Scattered radio frequency points of Internet of Things devices Develop joint frequency-domain and spatial-domain filtering: Spatial domain: Use beamforming technology to enhance the transformer direction gain; Frequency domain: Design a comb-shaped stopband filter to suppress communication carriers; Extract discharge signals greater than 5 pC under a background noise of -85 dBm.

[0039] 2. Distributed diagnosis of cable trenches Deploy 32 sensing nodes along the 10 kV cable trench, with a time-base synchronization accuracy of less than 1 ns; Adopt the pulse arrival time difference positioning method: ; Among them, is the time difference between the pulse arriving at sensors i and j, is the coordinate of the discharge source, is the electromagnetic wave propagation speed, and are the coordinates of sensors i and j; Draw a three-dimensional discharge hot spot cloud map.

[0040] 3. Digital twin linkage Map the real-time monitoring data to the BIM model; When a discharge in the cable joint is identified: Retrieve the historical load curve to evaluate the insulation aging rate; Combine the ground current data to diagnose the degree of sheath damage; Predict the remaining life and optimize the maintenance schedule; Generate a health record for the entire life cycle of the equipment.

[0041] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A discharge monitoring method based on dynamic frequency band selection and network identification, characterized in that, The method includes the following steps: Step S1: Collect the original electromagnetic signals of the power equipment through the UHF sensor array to generate an initial UHF signal dataset; Step S2: Input the initial UHF signal dataset into the multi-band filtering module for parallel band processing to generate a low-frequency band-pass signal, a high-frequency band-pass signal, a full-frequency direct communication signal, and a selective dual-band combined signal; Step S3: Respectively perform time-domain feature extraction, frequency-domain feature extraction, and statistical feature extraction on the signals of each frequency band output in Step S2 to generate a multi-dimensional feature vector set; Step S4: Input the multi-dimensional feature vector set into a pre-trained neural network classification model for signal validity verification, output the partial discharge signal probability value and the noise probability value, and generate signal classification identification data; Step S5: Calculate the real-time signal-to-noise ratio SNR of each frequency band based on the signal classification identification data, and combine it with a preset pulse density threshold to generate a frequency band performance evaluation matrix; Step S6: Execute dynamic frequency band switching decision-making according to the frequency band performance evaluation matrix; Step S7: Adjust the frequency band polling period according to the switching mode parameters configured by the user, generate an optimized partial discharge monitoring result, and output an alarm instruction.

2. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, wherein The said Step S2 includes: Step S21: Configure four parallel filtering links through the programmable filter module: Link 1: Use a 550 MHz band-pass filter to generate a low-frequency band-pass signal; Link 2: Use a 1300 MHz band-pass filter to generate a high-frequency band-pass signal; Link 3: Use a 900 MHz all-pass filter to generate a full-frequency direct communication signal; Link 4: Dynamically select two non-overlapping frequency bands for combined filtering to generate a dual-band combined signal; Step S22: Control the signal output timing of each filtering link through a multiplexer switch.

3. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, characterized in that The operation of frequency-domain feature extraction in the said Step S3 is specifically: Step S31: Perform discrete wavelet transform on each frequency band signal and decompose it into 8 sub-bands; Step S32: Calculate the energy proportion of each sub-band; ； Among them, is the energy proportion of the k-th sub-band, is the wavelet coefficient of the k-th sub-band, is the sampling point index, N is the number of sampling points, is the sub-band index; Step S33: Use the energy proportion distribution vector as the frequency-domain feature.

4. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, characterized in that The construction of the neural network classification model in the said Step S4 includes: Step S41: The model structure adopts a CNN-LSTM hybrid architecture: Input layer: Receive a feature vector with a dimension of 12; Convolution layer: 3 layers of 1D-CNN, with a convolution kernel size of 3×1; LSTM layer: The number of bidirectional LSTM units is 64; Output layer: The Softmax activation function outputs the partial discharge / noise probability; Step S42: The training data includes: Positive samples: Pulse waveforms of corona discharge and floating discharge; Negative samples: Mobile phone signals, arc noise, switch operation interference.

5. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, characterized in that, The logic of dynamic frequency band switching decision-making in the said Step S6 is: Step S61: Initialize the polling period T, and activate the signal acquisition of each frequency band in sequence; Step S62: Calculate the comprehensive performance score of each frequency band; ； Among them, is the comprehensive performance score of the i-th frequency band, is the partial discharge probability, is the current signal-to-noise ratio, is the switching time consumption, is the weight coefficient; Step S63, select the highest frequency band as the main monitoring channel.

6. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, characterized in that The switching mode parameter configuration in the said Step S7 includes: Emergency mode: Polling period T = 1 s, SNR drop threshold = 15%; Steady state mode: Polling period T = 15 min, SNR drop threshold = 25%.

7. The discharge monitoring method based on dynamic frequency band selection and network identification according to claim 1, characterized in that The method also includes an online learning mechanism: Step S71: When unrecognized signal types continuously appear, trigger the data augmentation module to generate synthetic samples; Step S72: Update the neural network classification model using an incremental learning algorithm: ； Among them, is the updated model parameter, is the model parameter before update, is the learning rate, is the gradient of the loss function L with respect to the parameter θ, is the prediction output of the model for the new sample and is the true label of the new sample.

8. The discharge monitoring method based on dynamic frequency band selection and network identification according to any one of claims 1-7, characterized in that The calculation formula for the signal-to-noise ratio SNR in step S5 is: ； Among them is the average power of partial discharge pulses is the noise power in the pulse-free section 9. A discharge monitoring system based on dynamic frequency band selection and network identification for implementing the method according to any one of claims 1-8, characterized in that The system includes: A signal acquisition and processing module, a dynamic frequency band optimization module, and an intelligent recognition and analysis module; The signal acquisition and processing module includes: A very high frequency (VHF) sensor array: Arranged at the pot insulator of the GIS device to collect 900 MHz electromagnetic signals; A signal conditioning unit: Comprising a preamplifier and a 24-bit ADC converter; The dynamic frequency band optimization module includes: A programmable multi-bandpass filter bank: Hardware implementation of four parallel filtering links; A frequency band switching controller: A real-time SNR evaluation and switching logic circuit based on FPGA; The intelligent recognition and analysis module includes: A feature extraction engine: Perform time-domain, frequency-domain, and statistical feature calculations; A neural network processor: Deploy a CNN-LSTM classification model; An online learning unit: Support incremental model updates.

10. The discharge monitoring system based on dynamic frequency band selection and network identification according to claim 9, characterized in that, The programmable multi-bandpass filter bank is designed using a microstrip line structure and includes: A low-frequency path: Center frequency 600 MHz, bandwidth ±200 MHz; A high-frequency path: Center frequency 1250 MHz, bandwidth ±250 MHz; A dual-frequency combination path: Switch the frequency band combination method through PIN diodes.

Images