Intelligent defect detection method based on power system

Abstract

The invention discloses an intelligent defect detection method based on a power system, and belongs to the technical field of power system detection and fault diagnosis, and the method comprises the steps: obtaining voltage waveform data and current waveform data collected by a power parameter sensor, and temperature data, humidity data and vibration data collected by an environment sensor, and generating a multi-modal data set; performing time domain feature extraction and frequency domain feature extraction on the multi-modal data set to generate a power equipment feature vector and an environment interference feature vector; performing matching analysis on the electrical equipment feature vector according to a preset defect feature baseline library to generate a preliminary defect judgment result; and compensating and correcting the initial defect judgment result based on the environment interference feature vector to generate a final defect detection report. According to the method, a collaborative technology of multi-modal data fusion acquisition, dynamic threshold matching and environment compensation correction is adopted, and accurate defect identification and adaptive optimization detection under complex working conditions can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system detection and fault diagnosis, and in particular to an intelligent defect detection method based on a power system. BACKGROUND

[0002] At present, power system equipment defect detection is a key technology to ensure the safe operation of the power grid. The traditional method mainly relies on threshold monitoring of basic electrical parameters such as voltage and current. With the development of smart grids, equipment defect detection needs to integrate multi-source sensor data and adapt to complex operating environments. Existing technical solutions usually use fixed threshold detection algorithms, collect electrical parameters through a single type of sensor, and combine a pre-set rule base to determine abnormalities. Some improved solutions introduce environmental parameter monitoring, but do not establish an effective interference compensation mechanism.

[0003] Existing technical means are mostly based on independent sensor networks to collect electrical parameters, use Fourier transform to extract harmonic features, and determine device abnormalities by manually setting thresholds. Some solutions attempt to combine environmental parameters such as temperature and humidity, but use a simple weighted superposition method to process multi-source data. The feature library update relies on human experience and lacks an automated optimization mechanism.

[0004] The existing technology has the following shortcomings: multi-source heterogeneous data lacks effective spatiotemporal synchronization processing methods, leading to incomplete feature extraction; fixed threshold detection has a high false alarm rate when load dynamics change; environmental interference and equipment defect features are not effectively decoupled; and the feature library update lags behind changes in the actual operating state of the equipment. SUMMARY

[0005] To solve the above problems, the present application provides an intelligent defect detection method based on a power system, which uses a collaborative technology of multi-modal data fusion collection, dynamic threshold matching, and environmental compensation correction to achieve accurate defect identification and adaptive optimization detection under complex conditions.

[0006] The above objectives can be achieved through the following solutions: An intelligent defect detection method based on a power system includes obtaining voltage waveform data and current waveform data collected by power parameter sensors, and temperature data, humidity data, and vibration data collected by environmental sensors to generate a multi-modal data set; performing time-domain feature extraction and frequency-domain feature extraction on the multi-modal data set to generate a power equipment feature vector and an environmental interference feature vector; performing matching analysis on the power equipment feature vector based on a pre-set defect feature baseline library to generate a preliminary defect determination result; and compensating and correcting the preliminary defect determination result based on the environmental interference feature vector to generate a final defect detection report.

[0007] Optionally, the generating the multi-modal data set comprises: alternately transmitting data of the power parameter sensor and the environmental sensor by using a preset time-division multiplexing transmission protocol to generate a synchronous time sequence data stream; performing difference compression processing on the synchronous time sequence data stream according to a preset power frequency reference value to generate a compressed data packet; and storing the compressed data packet in association with a preset device location identifier to generate the multi-modal data set.

[0008] Optionally, the generating the power equipment feature vector and the environmental interference feature vector comprises: separating voltage fluctuation features, current harmonic features and equipment vibration spectrum features from the multi-modal data set to generate the power equipment feature vector; and extracting temperature change rates, humidity gradient distributions and electromagnetic interference intensities from the multi-modal data set to generate the environmental interference feature vector.

[0009] Optionally, the generating the preliminary defect determination result comprises: obtaining a feature matching threshold, a historical defect waveform and a feature weight in the defect feature baseline library; obtaining a power grid load rate in real time, selecting a corresponding feature matching threshold from a preset threshold mapping table based on the power grid load rate to generate a dynamic detection threshold; performing similarity calculation on the power equipment feature vector and the historical defect waveform to generate a defect feature matching degree; and generating a preliminary defect determination result with a defect probability when the defect feature matching degree exceeds the dynamic detection threshold.

[0010] Optionally, the generating the dynamic detection threshold comprises: obtaining a preset load period division rule to identify whether a current power grid operation period is in a peak period, a flat period or a valley period; performing association mapping based on the current power grid operation period and a preset historical false alarm rate library to adjust the feature matching threshold to generate a period adaptive threshold; and performing secondary correction on the period adaptive threshold in combination with the electromagnetic interference intensity in the environmental interference feature vector to generate the dynamic detection threshold.

[0011] Optionally, the generating the final defect detection report comprises: calculating a temperature-humidity coupling influence coefficient according to the environmental interference feature vector; weighting and fusing the temperature-humidity coupling influence coefficient and the defect probability in the preliminary defect determination result to generate a corrected defect probability; and generating a final defect detection report according to the corrected defect probability and a preset defect level division standard.

[0012] Optionally, the method further comprises: generating an operation and maintenance instruction containing a defect location, a defect type and a maintenance priority according to the final defect detection report; obtaining a false alarm cause and the defect location and defect type in the field maintenance feedback data to generate a feature library optimization parameter; and adjusting the feature weight in the defect feature baseline library according to the feature library optimization parameter to generate an updated feature baseline library.

[0013] Optionally, the generating the feature library optimization parameter comprises: performing feature extraction based on the field maintenance feedback data to generate a new defect feature template; counting the false alarm frequency of the same device position in a preset period to generate a feature matching suppression coefficient; and combining the new defect feature template and the feature matching suppression coefficient as the feature library optimization parameter.

[0014] Optionally, the preset defect feature baseline library comprises: obtaining fault waveform data, environmental monitoring data and maintenance record data in historical defect cases to generate an original training data set; performing noise filtering and feature alignment processing on the original training data set to generate a standardized defect feature set; and performing clustering analysis on the standardized defect feature set according to the defect type to generate the defect feature baseline library.

[0015] Based on the same inventive concept, the application also provides an intelligent defect detection system based on a power system, which comprises: a data acquisition module, which is used to acquire voltage waveform data and current waveform data collected by a power parameter sensor, and temperature data, humidity data and vibration data collected by an environmental sensor, and generate a multi-modal data set; a feature extraction module, which is used to perform time domain feature extraction and frequency domain feature extraction on the multi-modal data set to generate a power equipment feature vector and an environmental interference feature vector; a defect analysis module, which is used to perform matching analysis on the power equipment feature vector according to a preset defect feature baseline library to generate a preliminary defect determination result; and a result correction module, which is used to perform compensation correction on the preliminary defect determination result based on the environmental interference feature vector to generate a final defect detection report.

[0016] Compared with the prior art, the application has the following advantages: 1. The application realizes multi-dimensional feature extraction of the equipment operation state through heterogeneous data fusion of power parameters and environmental parameters, overcomes the feature missing problem caused by traditional single sensor data source, and improves the completeness of defect feature identification.

[0017] 2. The application adopts a dual adjustment mechanism of dynamic threshold matching and environmental compensation, effectively eliminates the influence of power grid load fluctuation and external environmental interference on the detection result, and significantly improves the detection stability under complex working conditions.

[0018] 3. The application constructs a closed-loop optimization feedback loop, reversely corrects the feature library parameters with field maintenance data, so that the detection system has a continuous self-learning ability and adapts to the long-term operation requirements of equipment aging and environmental changes.

[0019] 4. The application cooperatively designs time-sharing multiplexing transmission and difference compression algorithm to reduce the data transmission bandwidth while maintaining the spatiotemporal correlation of features, and realizes the synchronous optimization of detection efficiency and data quality.

[0020] Other features and advantages of the present application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The purposes and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 is a flowchart of an intelligent defect detection method based on a power system according to an embodiment of the present application.

[0023] Figure 2 is a multi-modal data acquisition timing diagram according to an embodiment of the present application.

[0024] Figure 3 is a dynamic threshold adjustment mechanism diagram according to an embodiment of the present application.

[0025] Figure 4 is an environmental compensation correction mechanism diagram according to an embodiment of the present application.

[0026] Figure 5 is a structural diagram of an intelligent defect detection system based on a power system according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0028] With reference to Figure 1 , one embodiment of the present application proposes an intelligent defect detection method based on a power system, which adopts a cooperative technology of multi-modal data fusion acquisition, dynamic threshold matching and environmental compensation correction, and can realize accurate defect identification and adaptive optimization detection under complex working conditions.

[0029] The method according to the embodiment specifically includes: The voltage waveform data and the current waveform data collected by the power parameter sensor and the temperature data, the humidity data and the vibration data collected by the environment sensor are acquired to generate a multi-modal data set; Time domain feature extraction and frequency domain feature extraction are performed on the multi-modal data set to generate a power equipment feature vector and an environmental interference feature vector; The power equipment feature vector is matched and analyzed according to a preset defect feature baseline library to generate a preliminary defect determination result; The preliminary defect determination result is compensated and corrected based on the environmental interference feature vector to generate a final defect detection report.

[0030] Specifically, the method first collects voltage waveform data and current waveform data of the power equipment through the power parameter sensor, and collects temperature data, humidity data and vibration data through the environment sensor, and integrates these data into a multi-modal data set. Then, time domain feature extraction and frequency domain feature extraction are performed on the multi-modal data set. The time domain features include the amplitude, mean value and fluctuation range of voltage and current, and the frequency domain features are obtained by Fourier transform to obtain frequency spectrum components, thereby generating a power equipment feature vector and an environmental interference feature vector. Then, the power equipment feature vector is matched and analyzed with the preset defect feature baseline library to determine whether there is a defect by calculating the similarity or distance measure to generate a preliminary defect determination result. Finally, the preliminary result is compensated and corrected based on the environmental interference feature vector, for example, by weighting or threshold adjustment to eliminate the interference of environmental factors on the state of the power equipment, to generate a final defect detection report. This method improves the accuracy and robustness of defect detection through multi-modal data fusion and environmental interference compensation, effectively distinguishes real defects from environmental interference, avoids misjudgment, and improves the reliability of defect detection.

[0031] Optionally, the generating a multi-modal data set comprises: The data of the power parameter sensor and the environment sensor are alternately transmitted using a preset time division multiplexing transmission protocol to generate a synchronous time sequence data stream; The synchronous time sequence data stream is subjected to difference compression processing according to a preset power frequency reference value to generate a compressed data packet; The compressed data packet is stored in association with a preset device location identifier to generate a multi-modal data set.

[0032] Specifically, first, the communication interfaces of the power parameter sensors and the environmental sensors are configured. The power parameter sensors include voltage transformers and current transformers, and the environmental sensors include temperature sensors, humidity sensors, and vibration sensors. In the data transmission stage, a preset time-division multiplexing transmission protocol is used to control the sensors to alternately send data in a preset 0.1-second cycle. The power parameter sensors send voltage waveform data and current waveform data in odd time slices, and the environmental sensors send temperature data, humidity data, and vibration data in even time slices, to form a synchronous time sequence data stream with time stamp markers. The time-division multiplexing transmission protocol refers to a communication mode in which different device data is transmitted in a time sequence in the same communication channel. When performing difference compression processing on the synchronous time sequence data stream, a power frequency reference value of the power system is extracted. The power frequency reference value refers to the fundamental component amplitude corresponding to the rated frequency of the power system. For the voltage waveform data, the instantaneous difference between the current sampling value and the power frequency reference value is calculated, and for the environmental sensor data, the change amount difference between two adjacent sampling periods is calculated. The above difference data is packaged according to a preset compression encoding format to generate a compressed data packet. The compression encoding format uses a differential pulse code modulation method to convert the continuously changing analog quantity difference into a digital signal. Finally, the compressed data packet is stored in association with a preset device location identifier, which is a preset geographic coordinate encoding of the power device. In the storage process, a three-dimensional index relationship between the time stamp, the device location identifier, and the compressed data packet is established to form a multi-modal data set containing spatiotemporal correlation characteristics. As shown in FIG. 8, the figure shows the time sequence signals of the multi-modal data (voltage, current, temperature, humidity) collected in the power device monitoring over time from top to bottom, which shows that the multi-modal data collected by the system covers electrical parameters (voltage, current) and environmental parameters (temperature, humidity), providing a basis for subsequent analysis. Through the synergistic effect of the time-division multiplexing transmission protocol and the difference compression processing, the spatiotemporal consistency of the multi-source data is maintained while reducing the data transmission bandwidth. The time-division multiplexing technology avoids signal interference caused by concurrent transmission of multi-sensor data, and the difference compression algorithm effectively reduces the amount of redundant data by utilizing the power frequency characteristics of the power system and the continuity characteristics of the environmental parameters. The combination of the two produces a synergistic technical effect, which not only solves the data transmission congestion problem of traditional detection systems, but also provides a complete data basis for subsequent feature extraction, achieving simultaneous improvement of transmission efficiency and data quality. Figure 2

[0033] Optionally, the generating the power device feature vector and the environmental interference feature vector comprises: separating voltage fluctuation features, current harmonic features, and device vibration spectrum features from the multi-modal data set to generate a power device feature vector; extracting temperature change rates, humidity gradient distributions, and electromagnetic interference intensities from the multi-modal data set to generate an environmental interference feature vector. ​

[0034] Specifically, first extract the voltage waveform data and current waveform data from the multimodal data set, calculate the effective value of the voltage waveform and count the maximum and minimum values ​​within the preset time window to calculate the voltage fluctuation amplitude. ,have: , in, is the maximum voltage, is the minimum voltage, both of which are obtained by collecting data through voltage transformers. Perform fast Fourier transform on the current waveform, extract the amplitude of the fundamental component and each harmonic component, and calculate the total harmonic distortion rate of the current. ,have: , in, For the Subharmonic current effective value, is the effective value of the fundamental current, both of which are calculated from the original data collected by the current transformer. The current harmonic characteristics refer to the frequency components in the current signal that deviate from the fundamental frequency by integer multiples. To extract the vibration spectrum characteristics of the equipment, an acceleration sensor is used to collect the vibration signal, and the 10Hz-1000Hz frequency band signal is retained through a bandpass filter to calculate the vibration energy spectrum density. ,have: , in, is the Fourier transform result of the vibration signal, and is the preset vibration characteristic frequency band boundary value, Express Find the definite integral of the function, where the upper and lower limits of the integral are and The voltage fluctuation amplitude, current harmonic total distortion rate and vibration energy spectrum density are combined into the power equipment feature vector according to the preset weights. For the generation of environmental interference feature vectors, the temperature data sequence is extracted from the multimodal data set to calculate the temperature change rate per unit time. ,have: , in, is the temperature difference between adjacent sampling points, is the sampling time interval. The humidity gradient distribution is obtained by calculating the humidity difference at different monitoring points of the same equipment. ,have: , in, and is the measurement value of two adjacent humidity sensors. The electromagnetic interference intensity is calculated by the time domain signal collected by the magnetic field intensity sensor to calculate its root mean square value. ,have: , in, is the total number of sampling points, The first time domain signal collected by the magnetic field intensity sensor sampling values. The temperature change rate, humidity gradient distribution and electromagnetic interference intensity are combined into an environmental interference feature vector in a preset ratio. This method realizes the independent characterization of the power equipment state and environmental interference through the physical quantity feature separation technology. The voltage fluctuation characteristics reflect the equipment operation stability, the current harmonic characteristics characterize the electrical circuit abnormality, and the vibration spectrum characteristics indicate mechanical structure defects. The three are combined to form a comprehensive equipment state vector. At the same time, the temperature change rate reveals heat dissipation anomalies, the humidity gradient reflects the local condensation risk, and the electromagnetic interference intensity quantifies the external environmental disturbance. The three work together to build an environmental interference assessment system. Feature separation processing avoids the mixed operation of parameters of different dimensions, so that the equipment defect judgment is not affected by environmental noise. At the same time, the correlation between key features is retained, providing analyzable input parameters for subsequent compensation and correction, significantly improving the defect identification accuracy under complex working conditions.

[0035] Optionally, generating a preliminary defect determination result includes: Obtaining feature matching thresholds, historical defect waveforms, and feature weights from the defect feature baseline library; Acquire the grid load rate in real time, select the corresponding feature matching threshold from a preset threshold mapping table based on the grid load rate, and generate a dynamic detection threshold; Calculating similarity between the power equipment feature vector and the historical defect waveform to generate a defect feature matching degree; When the defect feature matching degree exceeds the dynamic detection threshold, a preliminary defect determination result with a defect probability is generated.

[0036] Specifically, first, the pre-stored feature matching threshold, historical defect waveform data and feature weight coefficient are obtained from the defect feature baseline library. These baseline data come from the long-term accumulated equipment failure case library. The defect feature baseline library contains feature waveform templates of various typical defect types. The load fluctuation index is calculated according to the real-time acquired power grid load rate. The load fluctuation index is calculated by the ratio of the current total power of the power grid to the reference power. The preset threshold mapping table is queried based on the load fluctuation index. The threshold mapping table stores the corresponding relationship between the load fluctuation range and the feature matching threshold. The corresponding feature matching threshold is selected as the dynamic detection threshold. When calculating the similarity of the power equipment feature vector, the vector space model is used to calculate the cosine similarity between the power equipment feature vector and the historical defect waveform feature vector in the defect feature baseline library. For the calculation of the defect feature matching degree , there are: , wherein, is the current power equipment feature vector, which includes normalized voltage fluctuation amplitude, current harmonic distortion rate and vibration energy spectrum density and other features; is the historical defect waveform feature vector, which comes from the defect feature baseline library and also includes normalized standard defect features; represents the dot product operation of two vectors; represents the modulus of the vector. The historical defect waveform feature vector is generated by the standardized defect feature set, which includes the standardized values of voltage fluctuation amplitude, current harmonic distortion rate and vibration energy spectrum density. When the calculated defect feature matching degree exceeds the dynamic detection threshold, it is determined that there is a device defect, and a preliminary defect judgment result including defect probability and similarity value is generated. Through the synergistic effect of the dynamic threshold adjustment mechanism and the feature matching algorithm, accurate judgment under the fluctuating load condition of the power grid is realized. The dynamic detection threshold is automatically adjusted according to the real-time load change, avoiding false judgments caused by fixed thresholds under load mutations. The defect feature matching degree calculation eliminates the dimensional difference of different dimensional feature parameters, ensuring the objectivity of the matching result. The combination of the two makes the defect detection system not only adapt to the dynamic changes of the power grid operating state, but also accurately identify complex defect features, solving the technical contradiction that the sensitivity and stability of traditional methods are difficult to balance under variable load conditions, and significantly improving the defect recognition reliability under complex conditions.

[0037] Optionally, the generation of the dynamic detection threshold comprises: acquiring a preset load period division rule to identify whether the current power grid operating period is in a peak period, a flat period or a valley period; associating and mapping the current power grid operating period with the preset historical false alarm rate library to adjust the feature matching threshold and generate a period adaptive threshold; The time period adaptive threshold is modified twice in combination with the electromagnetic interference intensity in the environmental interference feature vector to generate a dynamic detection threshold.

[0038] Specifically, first read the time period determination parameters from the preset load time period division rule. The load time period division rule divides the daily operation time period into peak time period, flat time period and valley time period according to the historical load curve of the power grid. Obtain the total power measurement value of the power grid at the current moment, and compare it with the preset time period power threshold range. When the total power exceeds the peak power threshold continuously for a preset time, it is determined to be a peak period. When it is lower than the valley power threshold, it is determined to be a valley period. The rest of the cases are determined to be flat periods. According to the identified current power grid operation time period, extract the false alarm rate statistics of the corresponding time period from the preset historical false alarm rate library. The historical false alarm rate library stores the mapping relationship between the feature matching threshold and the false alarm rate in each time period, and calculates the feature matching threshold adjustment amount that meets the target false alarm rate constraint by linear interpolation. For generating time period adaptive thresholds ,have: , in, The preset basic feature matching threshold is used as the reference value for adjustment; To adjust the sensitivity coefficient, a constant between 0.1 and 0.3 is preset to control the adjustment amplitude; It is the false alarm rate deviation coefficient of the current period, reflecting the degree to which the false alarm rate deviates from the target value. When the adaptive threshold of the period is corrected twice, the electromagnetic interference intensity is extracted from the environmental interference feature vector to correct the adaptive threshold. The electromagnetic interference intensity is obtained by calculating the root mean square value of the magnetic field sensor, and the environmental correction function is constructed to calculate the final dynamic detection threshold. ,have: , in, is the preset maximum allowable electromagnetic interference intensity. Exceed Time The environmental correction function quantifies the impact of electromagnetic interference on detection sensitivity into a threshold adjustment value, and finally generates a dynamic detection threshold. Figure 3As shown, the grid load rate is obtained in real time, and the shaded part in the figure is the peak period. As can be seen from the figure, the grid load rate increases significantly during the morning and evening peak periods (8-11 and 18-21), and the dynamic threshold value is adjusted according to the change of the load rate: the threshold value decreases when the load is high, and the detection sensitivity is improved. Through the dual adjustment mechanism of period characteristics and environmental interference, the intelligent optimization of the detection threshold value is realized. The period self-adaptive adjustment solves the problem of feature drift caused by load fluctuation, and the environmental interference correction eliminates the influence of external electromagnetic noise on the determination result. The synergistic effect of the two breaks through the technical limitations of the traditional fixed or single adjustment mode of the threshold value, so that the detection system can still maintain stable determination accuracy when the grid load periodically changes and sudden environmental interference coexists, while avoiding the risk of missed detection caused by excessive adjustment, and realizing the balanced improvement of detection sensitivity and anti-interference ability.

[0039] Optionally, the generating the final defect detection report comprises: calculating a temperature-humidity coupling influence coefficient according to the environmental interference feature vector; weighting and fusing the temperature-humidity coupling influence coefficient and the defect probability in the preliminary defect determination result to generate a corrected defect probability; generating a final defect detection report according to the corrected defect probability and a preset defect level division standard.

[0040] Specifically, first, the temperature change rate and the humidity gradient distribution are extracted from the environmental interference feature vector. The temperature change rate is calculated by dividing the temperature difference of adjacent sampling points by the sampling interval time, and the humidity gradient distribution is obtained from the humidity sensor measurement values of different monitoring points of the same device, so as to construct a temperature-humidity coupling influence coefficient calculation formula. For calculating the temperature-humidity coupling influence coefficient , , wherein, is a preset maximum allowed temperature change rate, which is determined based on historical data; is a preset maximum allowed humidity gradient value, which is also determined based on historical data. The temperature-humidity coupling influence coefficient represents the influence degree of the combined effect of environmental temperature and humidity on the insulation performance of the device. As shown in Figure 4 , the color depth represents the influence degree of environmental factors on defect detection, such as significant influence in high temperature and high humidity. The temperature-humidity coupling influence coefficient is weighted and fused with the defect probability in the preliminary defect determination result to generate a corrected defect probability. For calculating the corrected defect probability , , wherein, It is a preset environmental impact weight factor, which is set to a constant value between 0.1 and 0.5 according to the characteristics of the equipment insulation material; The defect probability in the initial preliminary defect determination result. According to the preset defect grade classification standard, the standard includes a mapping relationship between multiple defect probability intervals and corresponding defect grades. Compare with each interval threshold. When the first threshold is exceeded, it is determined to be a general defect, and when the second threshold is exceeded, it is determined to be a serious defect. At the same time, the vibration energy spectrum density value in the characteristic vector of the power equipment is combined to perform auxiliary verification of mechanical defects, and finally generate a final defect detection report containing the defect level, affected parts and recommended disposal measures. This method improves the reliability of the detection results through the quantification of environmental coupling effects and a multi-dimensional verification mechanism. The temperature-humidity coupling coefficient accurately reflects the accelerated effect of environmental factors on the aging of equipment insulation, and the weighted fusion algorithm eliminates the inflated or underestimated effects of environmental interference on the probability of defects. The dual judgment mechanism of defect level classification standard and vibration feature verification ensures that the detection conclusion meets the correlation requirements of electrical parameter anomalies and mechanical state changes at the same time, solves the misjudgment problem caused by a single judgment criterion in traditional methods, and significantly improves the defect location accuracy and the pertinence of disposal recommendations.

[0041] Optionally, the method further includes: Generate an operation and maintenance instruction including defect location, defect type and repair priority according to the final defect detection report; Obtain the false alarm causes, defect locations, and defect types from on-site maintenance feedback data, and generate feature library optimization parameters; The feature weights in the defect feature baseline library are adjusted according to the feature library optimization parameters to generate an updated feature baseline library.

[0042] Specifically, first match the equipment identifier according to the defect location information in the final defect detection report, and retrieve the standard maintenance process for the corresponding equipment type from the preset disposal solution library. Match the defect type code with the preset defect disposal mapping table to generate preliminary operation and maintenance instructions containing a list of maintenance items and operating specifications. Calculate the maintenance priority index by combining the defect probability value and defect level. ,have: , in, The weight coefficient corresponding to the defect level is determined by statistical analysis of the equipment failure history data, and finally generates the operation and maintenance instruction containing the defect position, defect type and maintenance priority. When obtaining the on-site maintenance feedback data, the defect type, defect position and false alarm reason classification code found in the actual maintenance process are input through the mobile terminal. The false alarm reason classification code includes three types of environmental interference misjudgment, sensor failure misjudgment and feature matching error. The corresponding power equipment feature vector and environmental interference feature vector are extracted for the false alarm case, and a false alarm feature data set is generated.

[0043] By comparing the difference between the false alarm feature data set and the actual situation, the feature library optimization parameter is calculated, which reflects the direction and amplitude of the adjustment of each feature weight. Finally, according to the feature library optimization parameter, the feature weight in the defect feature baseline library is iteratively updated, and the gradient descent algorithm is used to gradually adjust the weight proportion of each feature dimension in the similarity calculation, and an updated feature baseline library is generated. This method forms a closed-loop system from detection to feedback to optimization, and continuously improves the detection accuracy through continuous learning.

[0044] Optionally, the generating a feature library optimization parameter comprises: extracting features based on the on-site maintenance feedback data to generate a new defect feature template; statistically analyzing the false alarm frequency of the same device position in a preset period to generate a feature matching suppression coefficient; combining the new defect feature template and the feature matching suppression coefficient into the feature library optimization parameter.

[0045] Specifically, first, the defect waveform change feature is extracted from the on-site maintenance feedback data, which includes the actual defect type code and the false alarm reason classification code. The multi-modal data set corresponding to the real defect case is resampled to obtain the voltage waveform data, current waveform data and vibration signal data within a preset time window before the defect occurs. The mutation point amplitude and duration of the voltage waveform are extracted by wavelet transform, and the voltage sag feature parameter is calculated. For calculating the voltage sag feature parameter , , wherein, is the rated voltage value of the equipment, is the absolute value of the voltage mutation amplitude, is the voltage anomaly duration, is the logarithmic transformation of the duration to avoid the influence of extreme values. The harmonic component decomposition is performed on the current waveform data to extract the harmonic amplitude growth of a specific number, and the harmonic mutation index is calculated. For calculating the harmonic mutation index , , wherein, is the harmonic reference value of the device in normal state, which is obtained by historical normal operation data statistics. The voltage sag characteristic parameter, the harmonic mutation index and the vibration energy spectrum density change amount are combined to generate a new defect feature template. When the false alarm frequency of the same device position in a preset period is counted, all matching records of the device position in the last preset days are extracted from the feature matching record database, so as to calculate the feature matching suppression coefficient. For calculating the feature matching suppression coefficient , there is: , wherein, is the effective matching times, which is obtained by consistency verification of matching results and maintenance results; is the total matching times; is the continuous false alarm times, which is counted from the time series of false alarm events; is the decay coefficient, which is preset as a constant value between 0.05 and 0.2 according to the type of the device. When the new defect feature template and the feature matching suppression coefficient are combined as the feature library optimization parameters, the new defect feature template is additionally labeled with an environmental correlation degree tag, which is obtained by calculating the standard deviation of the environmental interference feature vector when the defect occurs. The feature matching suppression coefficient is stored in combination with the device position identifier to form an optimized parameter set with spatial distribution characteristics. The dynamic optimization of the feature library is realized through the dual mechanisms of defect feature evolution tracking and false alarm mode identification. The new defect feature template captures the gradual defect features generated in the device aging process, and the feature matching suppression coefficient automatically suppresses the regional false alarm mode caused by environmental changes. The combination of the two enables the feature baseline library to not only timely incorporate new defect features, but also effectively eliminate repetitive false judgments, forming the adaptive ability of the detection system to environmental changes and device degradation, and significantly improving the stability and generalization performance of the defect detection model in long-term operation.

[0046] Optionally, the preset defect feature baseline library comprises: Obtain fault waveform data, environmental monitoring data and maintenance record data in historical defect cases to generate an original training data set; Perform noise filtering and feature alignment processing on the original training data set to generate a standardized defect feature set; According to the defect type, the standardized defect feature set is subjected to cluster analysis to generate the defect feature baseline library.

[0047] Specifically, first, the fault waveform data, environmental monitoring data and repair record data are extracted from the historical defect case database, and the historical defect case database stores the real defect event data set verified by the field. The fault waveform data is preprocessed, and the sliding window filtering algorithm is used to eliminate high-frequency noise, and the filter window width is set to an integer multiple of the power frequency period. The effective value fluctuation sequence is extracted from the filtered voltage waveform data, the current waveform data is decomposed into fundamental and harmonic components, and the vibration signal data is converted into a time-frequency spectrum by short-time Fourier transform. When filtering out noise from the original training data set, a composite filtering function is constructed. For the output signal after composite filtering , there are: , wherein, is the output of the low-pass filter, used to eliminate high-frequency noise; is the output of the median filter, used to eliminate impulse noise; is the dynamic weight coefficient ( ), which is adjusted according to the signal signal-to-noise ratio to balance the effects of low-pass and median filtering. The composite filtering eliminates random interference while retaining feature mutation information, generating denoised waveform data. When performing feature alignment processing, a multi-source data time synchronization mechanism is established. The voltage waveform sampling points, current waveform sampling points and vibration signal sampling points are interpolated and resampled, and the time reference is unified to the power frequency period equal division time axis. After feature alignment, a standardized defect feature set is generated, including the aligned voltage effective value sequence, current harmonic content sequence and vibration energy distribution matrix. In the clustering analysis process, an improved density clustering algorithm is used to define a feature similarity measure function to calculate the comprehensive similarity between two defect features. For the calculation of comprehensive similarity , there are: , wherein, ， ， is a preset weight parameter used to adjust the importance of voltage, current and vibration features in clustering, satisfying ; is the voltage fluctuation mode difference, which quantifies the difference of the voltage effective value sequence; is the harmonic distribution similarity, which quantifies the difference of the current harmonic content sequence; The vibration spectrum correlation coefficient is used to quantify the difference of the vibration energy distribution matrix. The density clustering algorithm automatically identifies defect categories with similar waveform characteristics, generates feature clustering clusters grouped by defect type, and finally forms a hierarchical structure of the defect feature baseline library. This method constructs a high-precision feature library through multi-source data fusion processing and intelligent clustering technology. The composite filtering algorithm solves the problem of feature distortion caused by traditional single filtering method, the time synchronization mechanism ensures the spatio-temporal correlation of multi-physical quantity features, and the improved density clustering algorithm effectively identifies potential defect patterns. The implementation of the technical scheme makes the feature baseline library not only accurately represent the typical defect characteristics, but also have the inclusiveness of new defect patterns, providing a reliable feature comparison benchmark for subsequent defect detection, and significantly improving the accuracy and system scalability of defect type recognition.

[0048] Based on the same inventive concept, as shown in Figure 5 The application also provides an intelligent defect detection system based on a power system, which comprises: A data acquisition module is configured to acquire voltage waveform data and current waveform data collected by a power parameter sensor, and temperature data, humidity data and vibration data collected by an environmental sensor, and generate a multi-modal data set; A feature extraction module is configured to perform time domain feature extraction and frequency domain feature extraction on the multi-modal data set, and generate a power equipment feature vector and an environmental interference feature vector; A defect analysis module is configured to perform matching analysis on the power equipment feature vector according to a preset defect feature baseline library, and generate a preliminary defect determination result; A result correction module is configured to compensate and correct the preliminary defect determination result based on the environmental interference feature vector, and generate a final defect detection report.

[0049] In order to verify the feasibility of the application in implementation, the application is applied to a 110kV Chengnan substation in a certain urban area. The substation undertakes the power supply task of the key area, and the reliability requirement of the equipment operation is extremely high. The traditional manual inspection and offline monitoring method is difficult to find early defects in real time, and is easy to be disturbed by extreme weather and other environmental factors, resulting in misjudgment. The application is applied to the substation, aiming to improve the operation and maintenance efficiency and the stability of power grid operation through continuous and intelligent defect detection of power equipment.

[0050] In order to verify the effectiveness of the application, a group of cable terminals and supporting switch equipment of the substation are continuously monitored and tested for 6 months, and the monitoring system deploys the data acquisition module, feature analysis module, defect determination module, environmental compensation module and feedback optimization module of the application. During the monitoring process, the multi-dimensional data of the equipment under different power grid loads and different seasonal environments are recorded.

[0051] In the embodiment, the data acquisition module generates a synchronous time sequence data stream by deploying power parameter sensors (voltage transformers, current transformers) and environmental sensors (temperature, humidity, vibration sensors) near the cable terminal, using a time-division multiplexing transmission protocol. At 15:30 on August 15, 2024 (peak period of the power grid), the system obtains the voltage waveform, current waveform, and temperature 38℃, humidity 85%, equipment vibration data of the No. 1 cable terminal. After differential compression processing, these data are stored in association with the device location identifier "ZN-CB-01", forming a multi-modal data set.

[0052] The feature analysis module processes the multi-modal data set. From the power data, the voltage fluctuation amplitude is extracted as 3.5% of the rated value, the current harmonic total distortion rate THD is 4.8%, and the vibration energy spectrum density is significantly increased. These parameters are combined into a power equipment feature vector. At the same time, from the environmental data, the temperature change rate is extracted as 2℃ / h, and the humidity gradient β is high on the surface of the equipment, generating an environmental interference feature vector.

[0053] In the preliminary defect judgment stage, the defect judgment module selects a higher feature matching threshold from the defect feature baseline library according to the load at the peak period, and generates a dynamic detection threshold. The system performs cosine similarity calculation on the power equipment feature vector and the "cable joint insulation aging" defect waveform in the baseline library, and obtains a matching degree of 0.82. Since the matching degree exceeds the dynamic detection threshold 0.80, the system generates a preliminary defect judgment result, preliminarily judging that "No. 1 cable terminal has insulation defect risk".

[0054] Subsequently, the environmental compensation module is started. According to the environmental interference feature vector, the temperature-humidity coupling influence coefficient is calculated to be high, and the environmental compensation parameters are generated. The parameters are weighted and fused with the defect probability in the preliminary defect judgment result, generating a corrected defect probability of 0.88. According to the defect level division standard, the final defect is judged as "general defect", and the final defect detection report and operation and maintenance instruction containing the defect location, defect type (insulation aging) and maintenance priority (medium) are generated.

[0055] According to the instruction, the on-site operation and maintenance team of the present application carries out maintenance on the "ZN-CB-01" device the next day, and the feedback data confirms that the cable terminal has early insulation slight aging phenomenon, which is consistent with the system detection result. The actual defect type and waveform characteristics in the maintenance feedback data are obtained by the feedback optimization module, generating new defect feature templates and feature library optimization parameters. The system adjusts the feature weights in the defect feature baseline library according to these parameters, and includes this early and subtle defect feature into the library, completing the closed-loop optimization of the detection model.

[0056] From the 6-month data comparison, the intelligent detection system of the application has significant advantages in defect identification accuracy, environmental adaptability and system self-optimization capability. Taking defect detection in high temperature and high humidity weather in summer as an example, the system reduces the false positive rate caused by environmental factors by about 60% through environmental compensation correction; compared with the traditional method, the accuracy of defect identification is improved from 85% to 98%; the feedback optimization module makes the system improve the ability to discover new and early defects by 30% after running for 3 months.

[0057] Table 1 Defect detection event data record table

[0058] Table 2 Defect judgment and compensation correction data table

[0059] Table 3 System performance comparison and optimization feedback data table

[0060] From the data recorded in the above tables 1 to 3, it can be seen that the application has achieved remarkable results in the actual application of Chengnan substation.

[0061] Table 1 clearly shows how the system obtains raw data from multiple source sensors and accurately extracts power equipment features and environmental interference features for analysis, laying a data foundation for subsequent intelligent judgment.

[0062] Table 2 details the complete process of defect judgment. The system not only dynamically adjusts the detection threshold according to the power grid load to avoid misjudgment during peak periods, but more importantly, through the environmental compensation algorithm, it scientifically corrects the probability of insulation defects in high temperature and high humidity environment, making the final judgment result closer to the physical reality and significantly improving the reliability of detection.

[0063] The data comparison in table 3 directly proves the technical advantages of the application. Compared with the traditional method, the application has greatly improved in accuracy and anti-interference ability. More importantly, through the closed-loop self-learning mechanism formed by the feedback optimization module, the system can learn from actual maintenance cases and continuously improve the recognition ability of early and new defects, which shows that the application has evolution ability and can adapt to the long-term evolution of power equipment state.

[0064] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean direct connection of the line, and indirect connection mode can also be used as long as the purpose of the application is achieved. The above-mentioned is only an exemplary embodiment of the application, and cannot limit the scope of the application.

[0065] intended to encompass any and all embodiments of the application with equivalents as would be ascertained by those skilled in the art to which the application pertains. Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

Claims

1. An intelligent defect detection method based on power system, characterized in that: The method comprises: Acquire voltage waveform data and current waveform data collected by power parameter sensors, as well as temperature data, humidity data, and vibration data collected by environmental sensors, to generate a multimodal data set; Performing time domain feature extraction and frequency domain feature extraction on the multimodal data set to generate a power equipment feature vector and an environmental interference feature vector; Perform matching analysis on the power equipment feature vector according to a preset defect feature baseline library to generate a preliminary defect determination result; The preliminary defect determination result is compensated and corrected based on the environmental interference feature vector to generate a final defect detection report.

2. The intelligent fault detection method based on the power system according to claim 1, characterized in that: Generating a multimodal data set includes: Using a preset time-division multiplexing transmission protocol to alternately transmit data from the power parameter sensor and the environmental sensor to generate a synchronous time-series data stream; Performing differential compression processing on the synchronous time series data stream according to a preset power frequency reference value to generate a compressed data packet; The compressed data packet is associated with a preset device location identifier and stored to generate a multimodal data set.

3. The intelligent defect detection method based on the power system according to claim 1, characterized in that: Generating the power equipment characteristic vector and the environmental interference characteristic vector includes: Separating voltage fluctuation features, current harmonic features, and equipment vibration spectrum features from the multimodal data set to generate a power equipment feature vector; The temperature change rate, humidity gradient distribution and electromagnetic interference intensity are extracted from the multimodal data set to generate an environmental interference feature vector.

4. The intelligent defect detection method based on the power system according to claim 1, characterized in that: Generating a preliminary defect determination result includes: Obtaining feature matching thresholds, historical defect waveforms, and feature weights from the defect feature baseline library; Acquire the grid load rate in real time, select the corresponding feature matching threshold from a preset threshold mapping table based on the grid load rate, and generate a dynamic detection threshold; Calculating similarity between the power equipment feature vector and the historical defect waveform to generate a defect feature matching degree; When the defect feature matching degree exceeds the dynamic detection threshold, a preliminary defect determination result with a defect probability is generated.

5. The intelligent defect detection method based on the power system according to claim 4 is characterized in that: Generating a dynamic detection threshold comprises: Obtain the preset load period division rules to identify whether the current power grid operation period is in the peak period, flat period or valley period; Based on the correlation mapping between the current power grid operation period and a preset historical false alarm rate library, the feature matching threshold is adjusted to generate a time period adaptive threshold; The time period adaptive threshold is modified twice in combination with the electromagnetic interference intensity in the environmental interference feature vector to generate a dynamic detection threshold.

6. The intelligent defect detection method based on the power system according to claim 4 is characterized in that: Generating the final defect detection report includes: Calculating the temperature-humidity coupling influence coefficient according to the environmental interference characteristic vector; Performing weighted fusion of the temperature-humidity coupling influence coefficient and the defect probability in the preliminary defect determination result to generate a corrected defect probability; A final defect detection report is generated based on the corrected defect probability and the preset defect grade classification standard.

7. The intelligent defect detection method based on the power system according to claim 6, characterized in that: The method further comprises: Generate an operation and maintenance instruction including defect location, defect type and repair priority according to the final defect detection report; Obtain the false alarm causes, defect locations, and defect types from on-site maintenance feedback data, and generate feature library optimization parameters; The feature weights in the defect feature baseline library are adjusted according to the feature library optimization parameters to generate an updated feature baseline library.

8. The intelligent defect detection method based on the power system according to claim 7, characterized in that: The generated feature library optimization parameters include: Extract features based on the on-site maintenance feedback data to generate a new defect feature template; Count the frequency of false alarms at the same device location within a preset period and generate a feature matching suppression coefficient; The newly added defect feature template and the feature matching suppression coefficient are combined into the feature library optimization parameter.

9. The intelligent defect detection method based on the power system according to claim 1, characterized in that: The preset defect feature baseline library includes: Obtain fault waveform data, environmental monitoring data, and maintenance record data from historical defect cases to generate an original training dataset; Performing noise filtering and feature alignment processing on the original training data set to generate a standardized defect feature set; Cluster analysis is performed on the standardized defect feature set according to the defect type to generate the defect feature baseline library.

10. An intelligent defect detection system based on a power system, applied to an intelligent defect detection method based on a power system according to any one of claims 1 to 9, characterized in that: The system comprises: The data acquisition module is used to obtain the voltage waveform data and current waveform data collected by the power parameter sensor, as well as the temperature data, humidity data and vibration data collected by the environmental sensor, to generate a multimodal data set; A feature extraction module, configured to perform time domain feature extraction and frequency domain feature extraction on the multimodal data set to generate a power equipment feature vector and an environmental interference feature vector; A defect analysis module is used to perform matching analysis on the power equipment feature vector according to a preset defect feature baseline library to generate a preliminary defect determination result; A result correction module is used to compensate and correct the preliminary defect determination result based on the environmental interference feature vector to generate a final defect detection report.

Images