Intelligent identification system and method based on time-frequency feature fusion and adaptive classification

Abstract

The invention discloses an intelligent identification system and method based on time-frequency feature fusion and adaptive classification, and relates to the technical field of power equipment state monitoring. Comprising an ultrahigh frequency signal processing unit, a filter amplification unit, an analog-to-digital conversion unit, a digital signal preprocessing unit, a multi-dimensional feature extraction unit, a discharge phase analysis unit, a classification model optimization unit, an interference elimination unit, a defect type identification and diagnosis unit and a discharge source positioning unit. The ultrahigh frequency signal processing unit is used for collecting and transmitting partial discharge weak electromagnetic signals, and the filtering and amplifying unit is used for filtering, eliminating interference and amplifying signal amplitude. Through the ultrahigh frequency signal processing unit, the filtering and amplifying unit and the analog-to-digital conversion unit, weak electromagnetic signals are accurately collected, interference is efficiently filtered out, the signals are amplified and converted, the accuracy and effectiveness of the detection process are improved, and the misjudgment risk is reduced through pulse segmentation and normalization, multi-dimensional feature extraction and classification model establishment and optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment condition monitoring, and in particular to an intelligent recognition system and method based on time-frequency feature fusion and adaptive classification. Background Art

[0002] In modern power systems, the safe and stable operation of power equipment is of crucial importance. Partial discharge (PD), as a key indicator reflecting the insulation state of power equipment, its accurate detection and analysis are of great significance for ensuring the reliable operation of power systems. However, the detection and processing of PD signals face many challenges. When power equipment is operating, the electromagnetic signals generated by PD are extremely weak and are often submerged in the complex electromagnetic environmental noise. For example, in places such as high-voltage substations, various electrical equipment is densely arranged, and the electromagnetic interferences generated by different equipment are intertwined, making it extremely difficult to effectively extract PD signals. Although the ultra-high frequency (UHF) band (300 MHz - 3 GHz) can capture PD signals, the signals are weak and require high-sensitivity sensors and low-loss transmission methods. Otherwise, the signals are prone to distortion or loss, resulting in detection failure. In addition to the electromagnetic interference generated by the operation of the equipment itself, the power system is also affected by external communication signals, natural environmental noise, etc. There may be overlap between communication signals and PD signals in the frequency spectrum, and power frequency interference will also have a serious impact on detection. Traditional detection methods are difficult to accurately identify PD signals in such a complex interference environment, and are prone to false positives and missed detections, affecting the accuracy of equipment condition assessment. PD signals contain rich information about the insulation state of equipment, but it is not easy to accurately extract and analyze these features from them. The characteristic changes of signals in the time domain and frequency domain are complex and are related to factors such as equipment type and operating conditions. For example, the PD signal characteristics of different types of power transformers are different, and the PD signals of the same equipment under different aging degrees are also different. Therefore, multi-dimensional and refined feature extraction and analysis methods are needed to comprehensively and deeply explore signal features to achieve accurate assessment of the insulation state of equipment. Accurately judging the type of PD signal and the type of equipment defect is the key to formulating a reasonable maintenance strategy. However, due to the diversity and complexity of PD signals, existing classification models are difficult to meet the high-precision recognition requirements. The PD signal characteristics corresponding to different defect types may be similar, which is prone to classification errors. Moreover, with the expansion of the scale of the power system and the replacement of equipment, higher requirements are put forward for the adaptability and accuracy of the classification model, and continuous optimization and improvement are needed. It is necessary to design an intelligent recognition system and method based on time-frequency feature fusion and adaptive classification to solve the above-mentioned problems. Summary of the Invention

[0003] The objective of the present invention is to address the drawbacks existing in the prior art, and propose an intelligent recognition system and method based on time-frequency feature fusion and adaptive classification, so as to solve the problems of high difficulty in detecting weak signals, high requirements for signal feature analysis in complex interference environments, and urgent need for classification and recognition accuracy in the above technical solutions.

[0004] To achieve the above objective, the present invention is realized through the following technical solutions: An intelligent recognition system based on time-frequency feature fusion and adaptive classification, including a very high frequency (VHF) signal processing unit, the output end of the VHF signal processing unit is connected to the input end of a filtering and amplification unit, the output end of the filtering and amplification unit is connected to the input end of an analog-to-digital conversion unit, the output end of the analog-to-digital conversion unit is connected to the input end of a digital signal preprocessing unit, the output end of the digital signal preprocessing unit is connected to the input end of a multi-dimensional feature extraction unit, the output end of the multi-dimensional feature extraction unit is connected to the input end of a discharge phase analysis unit, the output end of the discharge phase analysis unit is connected to the input end of a classification model optimization unit, the output end of the classification model optimization unit is connected to the input end of an interference elimination unit, the output end of the interference elimination unit is connected to the input end of a defect type identification and diagnosis unit, and the output end of the defect type identification and diagnosis unit is connected to the input end of a discharge source location unit; The VHF signal processing unit is used to collect and transmit weak electromagnetic signals of partial discharge. The filtering and amplification unit is used to filter out interference and amplify the signal amplitude. The analog-to-digital conversion unit is used to convert analog electrical signals into digital signals according to a theorem. The digital signal preprocessing unit is used to denoise, smooth, and resample the signal. The multi-dimensional feature extraction unit is used to extract various types of features such as the time-frequency of the signal. The discharge phase analysis unit is used to draw a map to judge the severity of partial discharge. The classification model optimization unit is used to select the best, adjust parameters, and fuse to improve the recognition rate. The interference elimination unit is used to identify and suppress various interference signals. The defect type identification and diagnosis unit is used to input data to identify the defect type. The discharge source location unit is used to locate the position of the discharge source according to the time difference.

[0005] Furthermore, the output end of the digital signal preprocessing unit is also connected to the input end of an anomaly judgment unit, the output end of the anomaly judgment unit is connected to the input end of the discharge source location unit. The anomaly judgment unit is used to perform data judgment on the digital signals output by the digital signal preprocessing unit for partial discharge anomalies, and output the abnormal data to the discharge source location unit to locate the position of the discharge source according to the time difference.

[0006] Further, the multi-dimensional feature extraction unit includes a time-frequency analysis module, a statistical feature calculation module, and a phase feature extraction module. The time-frequency analysis module is used to analyze the change of signal frequency over time. The statistical feature calculation module is used to calculate quantization signals such as mean and variance. The phase feature extraction module is used to lock the phase and extract relevant features.

[0007] Further, the classification model optimization unit includes a parameter tuning module and a model fusion module. The parameter tuning module is used to traverse parameter combinations to find the best parameters. The model fusion module is used to combine multiple models to improve the recognition effect.

[0008] Further, the interference elimination unit includes an interference recognition module and an interference suppression module. The interference recognition module is used to identify the current signal interference with a classifier. The interference suppression module is used to suppress various types of interference using different methods.

[0009] Further, the intelligent recognition method based on time-frequency feature fusion and adaptive classification includes the following specific recognition methods: S1. Signal acquisition: Through the ultra-high frequency information processing unit, a high-sensitivity UHF sensor with a working frequency band of 300 MHz - 3 GHz is used. Its built-in high-gain antenna or high-coupling efficiency capacitive component accurately captures the weak electromagnetic signals generated by partial discharge in the space around the power equipment, and transmits the signals to the filter amplification unit through a low-loss coaxial cable for accurate acquisition and transmission to ensure that the signals are not distorted. S2. Filter amplification: The filter amplification unit suppresses low-frequency interference below 300 MHz and high-frequency clutter above 3 GHz, and effectively retains the effective frequency band of the ultra-high frequency partial discharge signal. Then, a low-noise amplifier is used to increase the signal amplitude. The noise coefficient of the amplifier is less than 2 dB, and the signal amplification factor can be flexibly adjusted between 5 - 50 times. The effective frequency band of the partial discharge signal is output to the analog-to-digital conversion unit to achieve efficient filter amplification and optimize the overall quality of the signal. S3. Analog-to-digital conversion: By using a high-precision analog-to-digital converter with 12 bits or more, the sampling frequency is set to be above 5 GHz according to the Nyquist sampling theorem. The effective frequency band of the filtered and amplified partial discharge signal is converted into a digital signal stored and transmitted in binary code form. The quantization error is less than 0.1%, and the digital signal is transmitted to the digital signal preprocessing unit for high-standard analog-to-digital conversion to reduce information loss. S4. Pulse segmentation and normalization: In the digital signal preprocessing unit stage, a sliding window-based detection method is used, where the window width is dynamically adjusted between 10 - 2000 ns. The starting point of the pulse is determined by setting an adaptive threshold to segment and extract independent pulse sequences. Then, these sequences are normalized to unify the signal amplitude to the range [-1, 1], ensuring comparability of different pulse amplitudes. The sequences are input into the multi-dimensional feature extraction unit for accurate segmentation and normalization to improve the precision of analysis. S5. Multi-dimensional feature extraction: In the multi-dimensional feature extraction unit, the time-frequency analysis module splices and fuses the frequency domain features obtained from multi-scale frequency domain analysis and the features after alignment with the time domain by dynamic time warping; the statistical feature calculation module extracts time domain features; the phase feature extraction module analyzes the relationship between the pulse and the power frequency phase and constructs a phase distribution model; meanwhile, the data of the phase feature extraction is transmitted to the discharge phase analysis unit, and based on the phase feature, a discharge phase map is drawn. The abscissa of the map is the power frequency phase, and the ordinate is the number of discharges or the discharge amount. The distribution law of the discharge points on the map is analyzed to judge the severity of partial discharge. Multi-dimensional features are extracted to comprehensively characterize the signal. S6. Classification model establishment and optimization: The data after time-frequency analysis, statistical feature calculation, and discharge phase analysis are transmitted to the classification model optimization unit. Through decision tree, random forest, and neural network algorithm models, the parameters are optimized and adjusted based on a large amount of historical data. At the same time, the classification ability of different types of pulses of the model is evaluated, and the classification effect is adjusted according to the evaluation results to optimize the classification model and improve the classification accuracy. S7. Interference signal rejection: The data in the classification model optimization unit is transmitted to the interference judgment unit, where it is classified and compared with the partial discharge data in the normal state. The data with a range error within the specified threshold is marked as normal data, and the clustering algorithm is used to determine the distribution range of normal data. When the new data deviates from this range by more than a certain threshold, it is determined as abnormal data. At the same time, the normal data and abnormal data are transmitted to the defect type identification and diagnosis unit to effectively reject interference and ensure data purity. S8. Classification and identification: The defect type identification and diagnosis unit uses the neural network algorithm model to extract features from the convolutional layer of the data and complete classification through the fully connected layer for abnormal data, and outputs the defect type. For normal data, the probability of partial discharge signal is calculated and analyzed through the forest algorithm model to judge the pulse type, for accurate classification and identification to clarify the equipment status. S9, Result Output and Alarm: After analyzing the normal data through the defect type identification and diagnosis unit, the pulse type and confidence level are displayed in real time. The confidence level is displayed accurate to two decimal places, and a configurable alarm threshold is set. When the partial discharge signal probability reaches or exceeds the threshold, an alarm is triggered. At the same time, the data triggering the alarm threshold is input into the partial discharge source location unit for the location of the partial discharge source. The abnormal data directly undergoes the location of the partial discharge source in the partial discharge source location unit, and the alarm is intuitively output to ensure the safety of the system.

[0010] Furthermore, in the S5 multi-dimensional feature extraction step, it includes time-domain feature extraction and frequency-domain feature extraction. The time-domain feature extraction calculates the time-domain features of the pulse rise time, the number of wave peaks, and waveform symmetry, and quantitatively describes the characteristics of the pulse in the time domain. The frequency-domain feature extraction uses the wavelet scattering network for multi-scale frequency-domain analysis to obtain frequency-domain features. After dimensionality reduction processing, the features are spliced and fused with the features aligned with the time domain through dynamic time warping (DTW). During the dimensionality reduction process, more than 95% of the key information is retained, and the time-frequency fusion extracts features to accurately and comprehensively analyze partial discharge, providing accurate data for partial discharge analysis.

[0011] Furthermore, in the S7 interference signal rejection step, it includes phase distribution clustering and dynamic amplitude threshold. The phase distribution clustering analyzes the relationship between the pulse and the power frequency phase to construct a phase distribution model. If the pulse phase concentration is greater than 90%, it is determined as communication interference and rejected. The dynamic amplitude threshold calculates the average value and standard deviation (σ) of the signal amplitude, and rejects abnormal pulses with amplitudes exceeding the average value ±3σ. Based on the phase amplitude, the interference is removed, improving the accuracy and stability of the diagnosis. Using the phase amplitude threshold, the interference signal is effectively removed.

[0012] In summary, the present invention provides a UHF partial discharge pulse intelligent recognition method and system based on time-frequency feature fusion and adaptive classification, having the following beneficial effects: 1. Through the UHF signal processing unit, the filtering and amplification unit, and the analog-to-digital conversion unit, weak electromagnetic signals are accurately collected, interference is efficiently filtered, amplified, and converted, providing a high-precision and low-error data basis for subsequent processing, and improving the accuracy and effectiveness of the detection process.

[0013] 2. Through the time-frequency analysis module, the statistical feature calculation module, and the phase feature extraction module in the multi-dimensional feature extraction unit, the partial discharge signal features are deeply mined from multiple angles in the time-frequency domain, a comprehensive feature vector is constructed, and the severity of the partial discharge is accurately judged, providing rich and accurate data support for subsequent analysis.

[0014] 3. Through pulse segmentation and normalization, multi-dimensional feature extraction, and establishment and optimization of the classification model, accurately segment and normalize the pulses, comprehensively extract and fuse signal features, optimize the model to improve classification accuracy, provide a reliable basis for equipment status assessment, and reduce the risk of misjudgment.

[0015] 4. Through the interference signal elimination, classification and identification, and discharge source location unit, effectively identify and eliminate interference signals, accurately classify defects and judge pulse types, quickly locate the discharge source, ensure the stability of the detection system, reduce the risk of equipment failure, and contribute to the safe and stable operation of the power system. Brief Description of the Drawings

[0016] Figure 1 It is a schematic structural diagram of the intelligent recognition system based on time-frequency feature fusion and adaptive classification of the present invention; Figure 2 It is a schematic structural diagram of the multi-dimensional feature extraction unit of the intelligent recognition system based on time-frequency feature fusion and adaptive classification of the present invention; Figure 3 It is a schematic structural diagram of the classification model optimization unit of the intelligent recognition system based on time-frequency feature fusion and adaptive classification of the present invention; Figure 4 It is a schematic structural diagram of the interference elimination unit of the intelligent recognition system based on time-frequency feature fusion and adaptive classification of the present invention; Figure 5 It is a schematic flow diagram of the intelligent recognition method based on time-frequency feature fusion and adaptive classification of the present invention. Detailed 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] Embodiment: Please refer to Figure 1 - Figure 4As shown in the figure, the present invention provides a technical solution: an intelligent recognition system based on time-frequency feature fusion and adaptive classification, including a UHF signal processing unit. The output end of the UHF signal processing unit is connected to the input end of a filtering and amplifying unit. The output end of the filtering and amplifying unit is connected to the input end of an analog-to-digital conversion unit. The output end of the analog-to-digital conversion unit is connected to the input end of a digital signal preprocessing unit. The output end of the digital signal preprocessing unit is connected to the input end of a multi-dimensional feature extraction unit. The output end of the multi-dimensional feature extraction unit is connected to the input end of a discharge phase analysis unit. The output end of the discharge phase analysis unit is connected to the input end of a classification model optimization unit. The output end of the classification model optimization unit is connected to the input end of an interference elimination unit. The output end of the interference elimination unit is connected to the input end of a defect type identification and diagnosis unit. The output end of the defect type identification and diagnosis unit is connected to the input end of a discharge source location unit; The UHF signal processing unit is used to collect and transmit weak electromagnetic signals of partial discharge. The filtering and amplifying unit is used to filter out interference and amplify the signal amplitude. The analog-to-digital conversion unit is used to convert the analog electrical signal into a digital signal according to the theorem. The digital signal preprocessing unit is used to denoise, smooth and resample the signal. The multi-dimensional feature extraction unit is used to extract various features such as the time-frequency of the signal. The discharge phase analysis unit is used to draw a spectrum to judge the severity of partial discharge. The classification model optimization unit is used to select the best parameters, adjust the parameters and fuse to improve the recognition rate. The interference elimination unit is used to identify and suppress various interference signals. The defect type identification and diagnosis unit is used to identify the defect type by inputting data. The discharge source location unit is used to locate the position of the discharge source according to the time difference.

[0019] Please refer to Figure 1 As shown in the figure, the output end of the digital signal preprocessing unit is also connected to the input end of an abnormality judgment unit. The output end of the abnormality judgment unit is connected to the input end of the discharge source location unit. The abnormality judgment unit is used to judge the abnormality of the digital signal output by the digital signal preprocessing unit and output the abnormal data to the discharge source location unit to locate the position of the discharge source according to the time difference.

[0020] Please refer to Figure 2 As shown in the figure, the multi-dimensional feature extraction unit includes a time-frequency analysis module, a statistical feature calculation module and a phase feature extraction module. The time-frequency analysis module is used to analyze the change of signal frequency with time. The statistical feature calculation module is used to calculate quantization signals such as mean and variance. The phase feature extraction module is used to lock the phase and extract relevant features.

[0021] Please refer to Figure 3 As shown in the figure, the classification model optimization unit includes a parameter tuning module and a model fusion module. The parameter tuning module is used to traverse parameter combinations to find the best parameters. The model fusion module is used to combine multiple models to improve the recognition effect.

[0022] Please refer toFigure 4 As shown, the interference elimination unit includes an interference recognition module and an interference suppression module. The interference recognition module is used to recognize the current signal interference with a classifier, and the interference suppression module is used to suppress various types of interference by different methods.

[0023] Please refer to Figure 1 - Figure 4 As shown, the intelligent recognition method based on time-frequency feature fusion and adaptive classification includes the following specific steps: S1. Signal acquisition: The ultra-high frequency information processing unit uses a high-sensitivity UHF sensor with a working frequency band of 300 MHz - 3 GHz. Its built-in high-gain antenna or high-coupling efficiency capacitive component accurately captures the weak electromagnetic signals generated by partial discharge in the space around the power equipment, and transmits the signals to the filter amplification unit through a low-loss coaxial cable. This can ensure that the extremely weak electromagnetic signals generated by partial discharge of power equipment are accurately captured. The capture accuracy at the picometer level does not miss any subtle changes. The signal attenuation rate of the low-loss coaxial cable is less than 5%, ensuring that the signal is not distorted during transmission, providing a true and reliable original signal for subsequent filtering, amplification and other processing, and greatly improving the accuracy and effectiveness of the entire detection process; S2. Filtering and amplification: The filter amplification unit suppresses low-frequency interference below 300 MHz and high-frequency clutter above 3 GHz, and effectively retains the effective frequency band of the ultra-high frequency partial discharge signal. Then, a low-noise amplifier is used to increase the signal amplitude. The noise factor of the amplifier is less than 2 dB, and the signal amplification multiple can be flexibly adjusted between 5 - 50 times. The effective frequency band of the partial discharge signal is output to the analog-to-digital conversion unit. The strong suppression ability of the band-pass filter for low-frequency and high-frequency interference (the suppression ability can reach more than 60 dB) effectively removes noise interference. The low-noise amplifier maintains low noise while increasing the signal amplitude, ensuring high-quality signals output to the analog-to-digital conversion unit. At the same time, the signal amplification multiple can be flexibly adjusted to adapt to different intensities of the original signal, laying a foundation for accurate analog-to-digital conversion and ensuring more accurate data for subsequent processing; S3. Analog-to-digital conversion: By using a high-precision analog-to-digital converter with more than 12 bits, the sampling frequency is set to more than 5 GHz according to the Nyquist sampling theorem, and the effective frequency band of the filtered and amplified partial discharge signal is converted into a digital signal stored and transmitted in binary coding form. The quantization error is less than 0.1%, and the digital signal is transmitted to the digital signal preprocessing unit. The high-resolution analog-to-digital converter and high sampling frequency enable the details of the analog signal to be accurately restored, with extremely small quantization error, minimizing information loss to the greatest extent, providing a high-precision data basis for subsequent digital signal preprocessing, enabling subsequent operations such as pulse segmentation and feature extraction to be based on more accurate data, and thus improving the reliability and detection accuracy of the entire detection system; S4. Pulse segmentation and normalization: In the digital signal preprocessing unit stage, a sliding window-based detection method is used. The window width is dynamically adjusted between 10 - 2000 ns. The starting point of the pulse is determined by setting an adaptive threshold to segment and extract independent pulse sequences. Then, these sequences are normalized to unify the signal amplitude to the interval [-1, 1], ensuring the comparability of different pulse amplitudes. The sequences are input into the multi-dimensional feature extraction unit. The dynamically adjusted sliding window and adaptive threshold can accurately segment the pulses, effectively avoiding analysis errors caused by pulse overlap. The normalization process unifies pulses with different amplitudes to the same interval, making the features of different pulses comparable during subsequent multi-dimensional feature extraction, improving the accuracy and stability of feature extraction, and providing strong support for accurately analyzing partial discharge signals; S5. Multi-dimensional feature extraction: In the multi-dimensional feature extraction unit, the time-frequency analysis module splices and fuses the frequency domain features obtained from multi-scale frequency domain analysis and the features after dynamic time warping alignment in the time domain; the statistical feature calculation module extracts time domain features; the phase feature extraction module analyzes the relationship between the pulse and the power frequency phase and constructs a phase distribution model; at the same time, the data of the phase feature extraction is transmitted to the discharge phase analysis unit. Based on the phase feature, a discharge phase map is drawn. The abscissa of the map is the power frequency phase, and the ordinate is the discharge times or discharge amount. The distribution law of the discharge points on the map is analyzed to judge the severity of partial discharge. The multi-dimensional feature extraction comprehensively and deeply excavates the signal features. The fused feature vector can more accurately represent the partial discharge signal. The fusion of time domain and frequency domain features, as well as the phase feature analysis and discharge phase map drawing, provide detailed information about the partial discharge signal from multiple angles, helping to more accurately judge the severity of partial discharge and providing richer and more accurate data support for the subsequent establishment of the classification model and equipment status assessment; S6. Classification model establishment and optimization: The data after time-frequency analysis, statistical feature calculation, and discharge phase analysis are transmitted to the classification model optimization unit. Through decision tree, random forest, and neural network algorithm models, the parameters are optimized and adjusted based on a large amount of historical data. At the same time, the classification ability of different types of pulses of the model is evaluated, and the classification effect is adjusted according to the evaluation results. The model parameters are optimized using a large amount of historical data, enabling the decision tree, random forest, and neural network algorithm models to better learn the feature patterns of different pulses such as partial discharge signals and noise. By continuously evaluating and adjusting the classification effect, the classification accuracy of the model for different types of pulses is improved, providing a reliable classification basis for accurately judging the equipment status subsequently and reducing the risk of misjudgment; S7. Interference signal rejection: The data in the classification model optimization unit is transmitted to the interference judgment unit and classified and compared with the partial discharge data in the normal state. The data with a range error within the specified threshold is marked as normal data, and the clustering algorithm is used to determine the distribution range of the normal data. When the new data deviates from this range by more than a certain threshold, it is determined as abnormal data. At the same time, the normal data and abnormal data are transmitted to the defect type identification and diagnosis unit. By comparing with the normal data and using the clustering algorithm to determine the normal range, interference signals can be effectively identified. After accurately rejecting the interference signals, the purity of the data input to the defect type identification and diagnosis unit is greatly improved, the influence of interference on the diagnosis result is reduced, the subsequent identification and diagnosis of the defect type are more accurate and reliable, and the stability and reliability of the entire detection system are improved; S8. Classification and identification: The defect type identification and diagnosis unit uses the neural network algorithm model for abnormal data, extracts features from the convolutional layer of the data and completes classification through the fully connected layer, and outputs the defect type. For normal data, the forest algorithm model is used for calculation and analysis, and then the partial discharge signal probability is output to judge the pulse type. Different algorithm models are used to process abnormal data and normal data, giving full play to the advantages of the neural network algorithm in feature extraction and classification and the ability of the forest algorithm in judging the partial discharge signal probability. Accurately output the defect type and judge the pulse type, providing clear and definite equipment status information for equipment maintenance personnel, helping them timely understand the equipment situation and take corresponding measures; S9. Result output and alarm: After the normal data is analyzed by the defect type identification and diagnosis unit, the pulse type and confidence level are displayed in real time. The confidence level is displayed accurate to two decimal places, and a configurable alarm threshold is set. When the partial discharge signal probability reaches or exceeds the threshold, an alarm is triggered. At the same time, the data triggering the alarm threshold is input to the discharge source location unit for discharge source location. The abnormal data directly undergoes discharge source location in the discharge source location unit. The pulse type and confidence level are displayed in real time, intuitively showing the detection results to the user, facilitating the user to quickly understand the equipment status. At the same time, the configurable alarm threshold can be flexibly set according to actual needs to trigger the alarm in a timely manner, helping the operation and maintenance personnel quickly respond to potential faults. Locating the discharge source for the data triggering the alarm helps the operation and maintenance personnel quickly locate the fault point, take targeted measures, reduce the equipment failure risk, and ensure the safe and stable operation of the power system.

[0024] In the S5 multi-dimensional feature extraction step, it includes time-domain feature extraction and frequency-domain feature extraction. The time-domain feature extraction calculates the time-domain features of the pulse rise time, the number of wave peaks, and the waveform symmetry, and quantitatively describes the characteristics of the pulse in the time domain. The frequency-domain feature extraction uses the wavelet scattering network for multi-scale frequency-domain analysis to obtain frequency-domain features. After dimensionality reduction processing, it is concatenated and fused with the features aligned with the time domain through dynamic time warping (DTW). During the dimensionality reduction process, more than 95% of the key information is retained, simplifying the data while retaining important features, and then it is concatenated and fused with the time-domain features after alignment through DTW, presenting the partial discharge signal comprehensively from multiple perspectives, providing a comprehensive and accurate data basis for subsequent accurately judging the severity of partial discharge, establishing a classification model, and evaluating the equipment status, and greatly improving the reliability and accuracy of the analysis.

[0025] In the S7 interference signal elimination step, it includes phase distribution clustering and dynamic amplitude threshold. Phase distribution clustering analyzes the relationship between the pulse and the power frequency phase to construct a phase distribution model. If the pulse phase concentration is greater than 90%, it is determined as communication interference and eliminated. The dynamic amplitude threshold calculates the average value and standard deviation (σ) of the signal amplitude, and eliminates abnormal pulses with amplitudes exceeding the average value ±3σ, which can significantly improve the purity of the data input to the defect type recognition and diagnosis unit, reduce the interference of interference on the diagnosis result, make the subsequent recognition and diagnosis of the defect type more accurate and reliable, ensure the stable operation of the detection system, and provide strong support for equipment fault judgment.

[0026] The above is only the preferred embodiment of the present invention, and it is not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent change equivalent embodiments and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments according to the technical essence of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. An intelligent recognition system based on time-frequency feature fusion and adaptive classification, including a very high frequency (VHF) signal processing unit, characterized in that: The output end of the UHF signal processing unit is connected to the input end of the filtering and amplifying unit, the output end of the filtering and amplifying unit is connected to the input end of the analog-to-digital conversion unit, the output end of the analog-to-digital conversion unit is connected to the input end of the digital signal preprocessing unit, the output end of the digital signal preprocessing unit is connected to the input end of the multi-dimensional feature extraction unit, the output end of the multi-dimensional feature extraction unit is connected to the input end of the discharge phase analysis unit, the output end of the discharge phase analysis unit is connected to the input end of the classification model optimization unit, the output end of the classification model optimization unit is connected to the input end of the interference elimination unit, the output end of the interference elimination unit is connected to the input end of the defect type identification and diagnosis unit, and the output end of the defect type identification and diagnosis unit is connected to the input end of the discharge source location unit; The UHF signal processing unit is used to collect and transmit weak electromagnetic signals of partial discharge. The filtering and amplifying unit is used to filter out interference and amplify the signal amplitude. The analog-to-digital conversion unit is used to convert analog electrical signals into digital signals according to the theorem. The digital signal preprocessing unit is used to denoise, smooth and resample the signal. The multi-dimensional feature extraction unit is used to extract various features such as time-frequency of the signal. The discharge phase analysis unit is used to draw a graph to judge the severity of partial discharge. The classification model optimization unit is used to select the best parameters, tune the parameters and fuse to improve the recognition rate. The interference elimination unit is used to identify and suppress various interference signals. The defect type identification and diagnosis unit is used to identify the defect type by inputting data. The discharge source location unit is used to locate the position of the discharge source according to the time difference.

2. The intelligent recognition system based on time-frequency feature fusion and adaptive classification according to claim 1, wherein: The output end of the digital signal preprocessing unit is also connected to the input end of the anomaly judgment unit. The output end of the anomaly judgment unit is connected to the input end of the discharge source location unit. The anomaly judgment unit is used to judge the anomaly of the digital signal output by the digital signal preprocessing unit and output the abnormal data to the discharge source location unit to locate the position of the discharge source according to the time difference.

3. The intelligent recognition system based on time-frequency feature fusion and adaptive classification according to claim 1, wherein: The multi-dimensional feature extraction unit includes a time-frequency analysis module, a statistical feature calculation module and a phase feature extraction module. The time-frequency analysis module is used to analyze the change of signal frequency with time. The statistical feature calculation module is used to calculate quantization signals such as mean and variance. The phase feature extraction module is used to lock the phase and extract relevant features.

4. The intelligent recognition system based on time-frequency feature fusion and adaptive classification according to claim 1, wherein: The classification model optimization unit includes a parameter tuning module and a model fusion module. The parameter tuning module is used to traverse parameter combinations to find the best parameters. The model fusion module is used to combine multiple models to improve the recognition effect.

5. The intelligent recognition system based on time-frequency feature fusion and adaptive classification according to claim 1, wherein: The interference elimination unit includes an interference identification module and an interference suppression module. The interference identification module is used to identify the current signal interference with a classifier. The interference suppression module is used to suppress various interferences by different methods.

6. An intelligent recognition method based on time-frequency feature fusion and adaptive classification, which uses the intelligent recognition system based on time-frequency feature fusion and adaptive classification according to any one of claims 1-5, and is characterized in that: It includes the following specific steps: S1. Signal acquisition: The high-sensitivity UHF sensor with a working frequency band of 300 MHz - 3 GHz in the ultra-high frequency information processing unit precisely captures the weak electromagnetic signals generated by partial discharge in the space around the power equipment through its built-in high-gain antenna or high-coupling efficiency capacitive component, and transmits the signals to the filtering and amplification unit through a low-loss coaxial cable; S2. Filtering and amplification: The filtering and amplification unit suppresses low-frequency interference below 300 MHz and high-frequency clutter above 3 GHz, effectively retains the effective frequency band of the ultra-high frequency partial discharge signal, and then uses a low-noise amplifier to increase the signal amplitude. The noise factor of the amplifier is less than 2 dB, and the signal amplification factor can be flexibly adjusted between 5 - 50 times. The effective frequency band of the partial discharge signal is output to the analog-to-digital conversion unit; S3. Analog-to-digital conversion: By using a high-precision analog-to-digital converter with 12 bits or more, the sampling frequency is set to be above 5 GHz according to the Nyquist sampling theorem, and the effective frequency band of the filtered and amplified partial discharge signal is converted into a digital signal stored and transmitted in binary coding form. The quantization error is less than 0.1%, and the digital signal is transmitted to the digital signal preprocessing unit; S4. Pulse segmentation and normalization: In the digital signal preprocessing unit stage, a detection method based on a sliding window is used. The window width is dynamically adjusted between 10 - 2000 ns. The starting point of the pulse is determined by setting an adaptive threshold to segment and extract an independent pulse sequence, and then these sequences are normalized to unify the signal amplitude to the [-1, 1] interval to ensure the comparability of different pulse amplitudes, and the sequences are input into the multi-dimensional feature extraction unit; S5. Multi-dimensional feature extraction: In the multi-dimensional feature extraction unit, the time-frequency analysis module splices and fuses the frequency domain features obtained from multi-scale frequency domain analysis and the features after dynamic time warping alignment in the time domain; The statistical feature calculation module extracts time domain features; The phase feature extraction module analyzes the relationship between the pulse and the power frequency phase and constructs a phase distribution model; at the same time, the data of the phase feature extraction is transmitted to the discharge phase analysis unit. Based on the phase feature, a discharge phase map is drawn. The abscissa of the map is the power frequency phase, and the ordinate is the discharge times or discharge amount. The distribution law of the discharge points on the map is analyzed to judge the severity of the partial discharge; S6. Classification model establishment and optimization: The data after time-frequency analysis, statistical feature calculation, and discharge phase analysis are transmitted to the classification model optimization unit. Through decision tree, random forest, and neural network algorithm models, the parameters are optimized and adjusted based on a large amount of historical data. At the same time, the classification ability of different types of pulses of the model is evaluated, and the classification effect is adjusted according to the evaluation results; S7. Interference signal elimination: The data in the classification model optimization unit is transmitted to the interference judgment unit, and is classified and compared with the partial discharge data in the normal state. The data within the specified threshold range of the range error is marked as normal data, and the clustering algorithm is used to determine the distribution range of the normal data. When the new data deviates from this range by more than a certain threshold, it is determined as abnormal data. At the same time, the normal data and abnormal data are transmitted to the defect type identification and diagnosis unit; S8. Classification and Recognition: The defect type recognition and diagnosis unit uses a neural network algorithm model for abnormal data to extract features from the convolutional layer of the data and complete classification through the fully connected layer, and outputs the defect type. For normal data, the partial discharge signal probability is calculated and analyzed through a forest algorithm model to determine the pulse type. S9. Result Output and Alarm: After the normal data is analyzed by the defect type recognition and diagnosis unit, the pulse type and confidence level are displayed in real time. The confidence level is displayed accurately to two decimal places, and a configurable alarm threshold is set. When the partial discharge signal probability reaches or exceeds the threshold, an alarm is triggered. At the same time, the data triggering the alarm threshold is input into the power source localization unit for power source localization. Abnormal data is directly input into the power source localization unit for power source localization.

7. The intelligent recognition method based on time-frequency feature fusion and adaptive classification according to claim 6, wherein: In the S5 multi-dimensional feature extraction step, it includes time-domain feature extraction and frequency-domain feature extraction. The time-domain feature extraction calculates the time-domain features of the pulse rise time, number of wave peaks, and waveform symmetry, and quantitatively describes the characteristics of the pulse in the time domain. The frequency-domain feature extraction uses a wavelet scattering network for multi-scale frequency-domain analysis to obtain frequency-domain features. After dimensionality reduction processing, the features are spliced and fused with the features aligned with the time domain through dynamic time warping (DTW). During the dimensionality reduction process, more than 95% of the key information is retained.

8. The intelligent recognition method based on time-frequency feature fusion and adaptive classification according to claim 7, wherein: In the S7 interference signal elimination step, it includes phase distribution clustering and dynamic amplitude threshold. The phase distribution clustering analyzes the relationship between the pulse and the power frequency phase to construct a phase distribution model. If the pulse phase concentration is greater than 90%, it is determined as communication interference and eliminated. The dynamic amplitude threshold calculates the average value and standard deviation (σ) of the signal amplitude, and eliminates abnormal pulses with amplitudes exceeding the average value ±3σ.

Images