A fault diagnosis system for an electric variable measuring insulator detection device

Abstract

The application relates to the technical field of power system monitoring and fault diagnosis, and particularly discloses a fault diagnosis system of an electric variable measurement insulator detection device. The system comprises a synchronous acquisition module, a multi-physical field feature extraction module, an adaptive fault diagnosis engine and a feedback module. By synchronously acquiring voltage harmonic, leakage current and temperature data, a characteristic vector coupled with electric-thermal influence is extracted, and two-stage collaborative diagnosis is carried out by using a dynamic threshold and Mahalanobis distance, so that high sensitivity, adaptive accurate identification and early warning of early insulator latent faults are realized. By constructing the synchronous data acquisition module and the multi-physical field feature extraction module, the two key influencing factors of the running voltage harmonic component and the environmental temperature are first quantified into the diagnostic system in the form of features.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and fault diagnosis technology, specifically relating to a fault diagnosis system for an insulator detection device for measuring electrical variables. Background Technology

[0002] In the field of condition monitoring and fault diagnosis of power transmission and transformation equipment, reliability assessment of key external insulation components such as insulators is crucial, as it directly affects the safe and stable operation of the power grid. Among these technologies, online insulator monitoring based on electrical variable measurement aims to assess the insulation condition and provide early warnings of potential faults by collecting electrical signals from insulators during operation. The basic principle of this technology is that when the insulator surface is contaminated, aged, or develops defects such as cracks, its electrical characteristics, such as leakage current and voltage distribution, will undergo measurable changes.

[0003] Existing technologies typically use a single leakage current amplitude or pulse count as a fault criterion. However, this method has significant limitations: it ignores the modulation effect of operating voltage harmonic components on leakage current characteristics, and fails to consider the profound impact of ambient temperature changes on the conductivity of insulating materials and the process of surface contamination and moisture absorption.

[0004] This diagnostic model, based on a single parameter and detached from the complex operating environment, lacks sufficient sensitivity to progressive faults such as early pollution accumulation and localized arc development, making it prone to missed detections. Especially on actual lines with complex pollution composition and variable weather conditions, the false alarm rate and missed detection rate of threshold alarm mechanisms relying solely on leakage current remain high, failing to meet the urgent need for accurate and early identification of latent faults in insulators. Summary of the Invention

[0005] The purpose of this invention is to provide a fault diagnosis system for an insulator detection device for measuring electrical variables, in order to solve the problem that the existing technology, due to the neglect of the influence of operating voltage harmonic modulation and ambient temperature, results in insufficient sensitivity, high false alarm and false alarm rates in the diagnosis method based on a single leakage current parameter, and the inability to accurately identify early latent faults in insulators.

[0006] The technical solution of this invention is a fault diagnosis system for an insulator detection device for measuring electrical variables. This system constructs a closed-loop diagnostic architecture integrating multi-physics field coupled sensing, multi-dimensional feature fusion, and dynamic threshold collaborative decision-making. The system includes a synchronous data acquisition module, a multi-physics field feature extraction module, an adaptive fault diagnosis engine, and a diagnostic result output and feedback module.

[0007] The synchronous data acquisition module is used to achieve high-precision time-synchronized data acquisition from the electrical variable measurement sensor group deployed on the target insulator string and environmental sensors. Specifically, this module includes a voltage signal acquisition unit, a leakage current signal acquisition unit, and an ambient temperature acquisition unit. The voltage signal acquisition unit is used to acquire the power frequency voltage at the insulator operating point and its 2nd to 13th harmonic components, with a sampling frequency of not less than 10 kHz.

[0008] The leakage current signal acquisition unit is used to synchronously acquire the full waveform leakage current signal of the insulator, with a measurement bandwidth covering 0 Hz to 1 MHz and a dynamic range of no less than 120 dB. The ambient temperature acquisition unit is used to acquire ambient temperature data at the insulator installation point in real time, with a measurement accuracy better than ±0.5 degrees Celsius. The above three acquisition units are synchronized through a high-precision hardware clock source to ensure that all data samples have a unified timestamp, with a time synchronization error of less than 1 microsecond.

[0009] The multiphysics feature extraction module is connected to the synchronous data acquisition module. It is used to preprocess the synchronously acquired raw data and extract feature vectors that characterize the insulator state and are coupled with the influence of the electro-thermal environment. This module includes a signal preprocessing submodule, a harmonic modulation analysis submodule, a temperature compensation submodule, and a feature fusion submodule.

[0010] The signal preprocessing submodule first applies a digital notch filter to the original leakage current signal at the 50 Hz power frequency and its integer multiples of harmonics to suppress system background interference. Then, it performs a 5-point linear moving average on the filtered signal to smooth random noise. The harmonic modulation analysis submodule receives the preprocessed leakage current signal and voltage harmonic component data, and performs the following analysis: it calculates the total effective value of the leakage current signal in the 2 kHz to 20 kHz frequency band, denoted as the high-frequency leakage current component; it calculates the cross-correlation function between this high-frequency leakage current component and the amplitudes of the 3rd, 5th, and 7th harmonics of the voltage, obtaining three cross-correlation peaks as characteristic quantities representing the intensity of harmonic modulation on the leakage current.

[0011] The temperature compensation submodule receives ambient temperature data and constructs a temperature compensation function based on the Arrhenius model of the conductivity of insulating materials. This function uses 25 degrees Celsius as a reference and converts the fundamental component of the leakage current at the current temperature to its equivalent value at the reference temperature to eliminate the influence of temperature on the conductivity of the insulating material. The feature fusion submodule combines six features—the three cross-correlation peaks output by the harmonic modulation analysis submodule, the equivalent value of the fundamental leakage current after temperature compensation output by the temperature compensation submodule, and the pulse count and waveform kurtosis extracted directly from the preprocessed leakage current signal—into a 6-dimensional feature vector, which is then output to the adaptive fault diagnosis engine.

[0012] The adaptive fault diagnosis engine, the core decision-making unit of the system, is connected to the multiphysics feature extraction module. It performs dynamic threshold comparison and state classification based on the input multi-dimensional feature vector. This engine incorporates a dynamic threshold update model and a fault state discriminator based on Mahalanobis distance. The dynamic threshold update model generates alarm thresholds for each feature in the feature vector that are adaptive to its historical operating state and ambient temperature.

[0013] The model operates as follows: During the initial 30 days of system commissioning, feature vector data of insulators under sunny and dry weather are continuously collected and regarded as the health status benchmark dataset; for each feature in the benchmark dataset, its mean and three times the standard deviation are calculated as the initial static threshold of that feature; after the system enters normal operation, the initial static threshold is corrected by linear regression every 24 hours based on the distribution of ambient temperature data over the past 7 days, generating a dynamic prediction threshold for the next 24 hours. The correction coefficient is obtained by looking up a table of the temperature-feature relationship surface trained offline.

[0014] The fault state discriminator based on Mahalanobis distance receives the current feature vector and the dynamic thresholds for each feature quantity provided by the dynamic threshold update model, and performs two levels of discrimination: Level 1 discrimination compares the measured value of each feature quantity in the current feature vector with its corresponding dynamic threshold. If any two or more feature quantities exceed the limit simultaneously, a Level 1 alarm is immediately triggered, indicating a severe fault state. If only one feature quantity exceeds the limit or none exceed the limit, the process proceeds to Level 2 discrimination. Level 2 discrimination calculates the Mahalanobis distance of the current feature vector relative to the health state benchmark dataset. Two distance thresholds are set: a warning threshold and an alarm threshold. If the Mahalanobis distance is greater than the warning threshold but less than the alarm threshold, it is determined to be an early abnormal state, and a warning signal is generated. If the Mahalanobis distance is greater than or equal to the alarm threshold, it is determined to be a developing fault state, and a Level 2 alarm signal is generated. If the Mahalanobis distance is less than the warning threshold, it is determined to be a normal state.

[0015] The diagnostic result output and feedback module is connected to the adaptive fault diagnosis engine. It encapsulates information such as diagnostic status, specific out-of-limit features, Mahalanobis distance, and ambient temperature into a standard-format diagnostic report and uploads it to the remote monitoring master station via wired or wireless communication interfaces. Simultaneously, this module sends the feature vector and final status label of each diagnosis back to the health status benchmark dataset in the adaptive fault diagnosis engine. This dataset is used to initiate a rolling update algorithm after the system has run for 90 days, gradually replacing the initial benchmark data with the latest 90-day normal status data, enabling the system's diagnostic benchmark to adapt to the slow aging process of the insulators.

[0016] As one embodiment of the present invention, the calculation of the cross-correlation function in the harmonic modulation analysis submodule is achieved by using the fast Fourier transform method with Hanning window, the window length is 1024 sampling points and the overlap rate is 50%, to ensure the real-time performance and accuracy of the calculation.

[0017] As one embodiment of the present invention, the Arrhenius model compensation function in the temperature compensation submodule is specifically expressed as: [The function is defined as follows: the current temperature...] The fundamental effective value of leakage current measured at Celsius Through formula Equivalent value at 25 degrees Celsius ;in The activation energy of insulating materials, This is the Boltzmann constant, and its specific value is obtained by calibrating it through temperature rise experiments on samples of the same type of insulator.

[0018] As one embodiment of the present invention, the specific process of linear regression correction in the dynamic threshold update model of the adaptive fault diagnosis engine is as follows: using ambient temperature as the independent variable and the historical average value of the feature quantity as the dependent variable, the temperature-feature quantity slope k is obtained by least squares fitting; then the future temperature prediction value... Dynamic threshold ,in The initial static threshold, The base temperature is 25 degrees Celsius.

[0019] As one embodiment of the present invention, the rolling update algorithm in the diagnostic result output and feedback module has the following execution strategy: for each new diagnostic record that is determined to be in a normal state, its corresponding feature vector is added to the rolling cache queue, and the oldest record in the queue is removed; when the cache queue reaches a capacity of 90 days, every 30 days, the mean vector and covariance matrix of the health status benchmark dataset are recalculated using all the data in the current queue, and the benchmark parameters required for Mahalanobis distance calculation are updated accordingly.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. This invention, by constructing a synchronous data acquisition module and a multi-physics feature extraction module, for the first time incorporates two key influencing factors—operating voltage harmonic components and ambient temperature—into a diagnostic system in the form of quantifiable features. The cross-correlation features extracted by the harmonic modulation analysis submodule directly reflect the coupling mechanism between abnormal discharge activity and the power grid harmonic environment, exhibiting high sensitivity to early local arcing. The temperature compensation submodule eliminates the interference of temperature changes on the basic leakage current, enabling subsequent diagnosis based on temperature-normalized electrical characteristics, significantly improving the consistency of diagnostic results under different climatic conditions. This multi-physics coupled sensing mechanism fundamentally overcomes the shortcomings of traditional single-parameter methods in terms of insufficient information dimensions.

[0022] 2. The adaptive fault diagnosis engine designed in this invention adopts a two-level decision architecture combining dynamic thresholds and multivariate statistical discrimination. The dynamic threshold update model enables the alarm threshold to adaptively adjust with ambient temperature, avoiding false alarms caused by fixed thresholds at extreme temperatures. The second-level discrimination based on Mahalanobis distance comprehensively utilizes the overall statistical characteristics of the 6-dimensional feature vector, enabling it to keenly detect early, subtle anomalies where multiple feature quantities are within limits but the overall pattern has deviated from a healthy state, thus achieving early detection of latent faults. The two-level discrimination logic balances rapid response to sudden severe faults with sensitive detection of progressive faults.

[0023] 3. This invention achieves the system's self-evolution capability through a diagnostic result output and feedback module. The rolling update algorithm enables the health status benchmark to dynamically adjust along with the slow aging of the insulator itself, preventing the system from misjudging normal aging as a fault due to outdated benchmarks, and ensuring the long-term applicability and accuracy of the diagnostic model throughout the entire equipment lifecycle. This system constitutes a complete closed loop from data perception, feature extraction, intelligent diagnosis to model self-updating, providing a highly reliable, adaptive, and early warning-capable systematic solution for insulator condition assessment. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall technical architecture of the fault diagnosis system for the electrical variable measurement insulator detection device proposed in this invention;

[0025] Figure 2 This is a schematic diagram of the core principle framework of the adaptive fault diagnosis engine in this invention;

[0026] Figure 3 This is a flowchart of the multiphysics feature extraction module in this invention.

[0027] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the synchronous data acquisition and multi-physics feature extraction module in this invention;

[0028] Figure 5 This is a closed-loop update logic framework diagram of the diagnostic result output and feedback module in this invention. Detailed Implementation

[0029] Please refer to Figures 1 to 5 This invention provides a fault diagnosis system for an insulator detection device for measuring electrical variable. Please refer to the appendix. Figure 1 This system constructs a closed-loop diagnostic architecture that integrates multi-physics field coupled sensing, multi-dimensional feature fusion, and dynamic threshold collaborative decision-making. The system includes a synchronous data acquisition module, a multi-physics field feature extraction module, an adaptive fault diagnosis engine, and a diagnostic result output and feedback module. These modules are connected and interact via a high-speed data bus and precise timing control logic, forming a complete technology chain from raw signal perception to advanced intelligent diagnosis and model self-optimization.

[0030] The synchronous data acquisition module is the source of the system's physical world perception. This module is used to achieve high-precision time-synchronized data acquisition from the electrical variable measurement sensor group deployed on the target insulator string and environmental sensors. Please refer to the attached document. Figure 4 This module specifically includes a voltage signal acquisition unit, a leakage current signal acquisition unit, and an ambient temperature acquisition unit. These three units can be physically integrated into a field acquisition terminal with a protection level of not less than IP65. The terminal is fixed near the insulator string being tested via an insulated mounting bracket and connected to the sensor via a shielded cable.

[0031] The voltage signal acquisition unit is used to acquire the power frequency voltage and its 2nd to 13th harmonic components at the insulator operating point. The core of this unit is a high-precision voltage transformer, whose primary side is connected to the transmission line via capacitive voltage division, and whose secondary side output signal is connected to a 24-bit analog-to-digital converter (ADC). The sampling frequency of this ADC is configured to be no less than 10 kHz, meaning it acquires 10,000 voltage sample points per second.

[0032] The sampling process is driven by a master clock signal provided by a cryogenic crystal oscillator, whose frequency stability is better than one part per million. At the beginning of each sampling period, the analog-to-digital converter receives a global synchronization pulse to ensure that its sampling time is strictly aligned with other acquisition units in the system. The acquired raw voltage digital sequence is first fed into a digital anti-aliasing filter with a cutoff frequency set to 5 kHz to eliminate noise above the Nyquist frequency. The filtered data is temporarily stored in a first-in-first-out buffer with a depth of 4096 sample points, awaiting subsequent processing.

[0033] The leakage current signal acquisition unit is used to synchronously acquire the full-waveform leakage current signal of the insulator. This unit includes a through-hole Rogowski coil sensor, which is mounted on the grounding lead at the insulator's steel cap. The output of the Rogowski coil is connected to a transimpedance amplifier, converting the weak current signal into a voltage signal. This voltage signal is then fed into a dedicated current measurement analog-to-digital converter (ADC) with a measurement bandwidth covering 0 Hz to 1 MHz and a dynamic range of at least 120 dB. This ADC also receives synchronization pulses from a high-precision hardware clock source; its sampling clock is sourced from the same voltage signal acquisition unit and is phase-locked. To achieve the 120 dB dynamic range, the ADC employs oversampling and digital noise reduction techniques, achieving an effective bit depth of 22 bits. The acquired raw leakage current data is also processed by a programmable anti-aliasing filter and stored in an independent first-in-first-out buffer. The leakage current signal range can be preset according to the insulator type, with a typical value of 0 to 10 mA.

[0034] The ambient temperature acquisition unit is used to collect ambient temperature data at the insulator installation points in real time. This unit uses a platinum resistance temperature sensor with a measurement accuracy better than ±0.5 degrees Celsius. The sensor is installed inside a radiation shield to reduce measurement errors caused by direct sunlight. The analog output signal of the temperature sensor is digitized by a 16-bit resolution analog-to-digital converter, and the sampling frequency is typically set to 1 Hz, or once per second. Although the sampling rate is low, its sampling trigger signal also originates from the system's high-precision hardware clock source, ensuring that each temperature sample value is assigned a precise timestamp in the same time coordinate system as the voltage and current data.

[0035] The three acquisition units are synchronized using a unified high-precision hardware clock source. This clock source is typically a temperature-compensated crystal oscillator (TCC) or a temperature-controlled crystal oscillator (TCC) module with an output frequency of 10 MHz. This frequency signal is distributed to the clock management chip within each acquisition unit. The clock management chip generates a sampling clock for its respective analog-to-digital converter based on a preset frequency division factor.

[0036] Simultaneously, a periodic synchronization pulse signal, such as one pulse per second, emitted by the main control unit, is broadcast to all acquisition units via a dedicated synchronization signal line. Upon receiving the rising edge of the synchronization pulse, each acquisition unit resets the phase of its internal sampling counter and records that moment as the reference point for the data timestamp. This hardware synchronization mechanism ensures that all data samples acquired from the voltage signal acquisition unit, leakage current signal acquisition unit, and ambient temperature acquisition unit have a unified timestamp, keeping the time synchronization error of the entire system within less than 1 microsecond. The acquired raw data packets, containing timestamps, channel identifiers, and sampled values, are transmitted in real-time to the multiphysics feature extraction module via a high-speed serial peripheral interface bus.

[0037] The multiphysics feature extraction module is connected to the synchronous data acquisition module. It is used to preprocess the synchronously acquired raw data and extract feature vectors that characterize the insulator state and are coupled with the effects of the electro-thermal environment. Please refer to the appendix. Figure 3 With appendix Figure 4 This module can be implemented in hardware using an embedded digital signal processor or a field-programmable gate array (FPGA). Its software logic includes a signal preprocessing submodule, a harmonic modulation analysis submodule, a temperature compensation submodule, and a feature fusion submodule. The module processes data in frames, with each frame corresponding to a fixed analysis time window, such as 1 second.

[0038] The signal preprocessing submodule first receives the raw leakage current digital sequence from the synchronous data acquisition module. The first step of this submodule is to perform digital notch filtering to suppress strong background interference from the 50 Hz power frequency and its integer harmonics. The notch filter employs a second-order infinite impulse response structure, and its transfer function, through zero-pole configuration design, forms stopbands with a depth of no less than 40 dB at frequencies such as 50 Hz, 100 Hz, and 150 Hz. The filter coefficients are fine-tuned according to the actual power frequency of the system, which can be calibrated in real-time by the fundamental frequency provided by the voltage signal acquisition unit. After notch filtering, the power frequency components in the signal are greatly weakened, while retaining the high-frequency and pulse components reflecting surface leakage and internal defects of the insulator.

[0039] The second step is to smooth the filtered signal to suppress random white noise. Here, a 5-point linear moving average filter is used, and its mathematical operation is as follows: for the nth... Data points Its output value This operation is efficiently implemented in a digital signal processor using a circular buffer and an accumulation operation. The preprocessed leakage current signal is labeled as... The data is then output to the harmonic modulation analysis submodule and the feature fusion submodule.

[0040] The harmonic modulation analysis submodule also receives the pre-processed leakage current signal. And voltage harmonic component data acquired from the voltage signal acquisition unit. The core task of this submodule is to analyze the modulation effect of power grid harmonics on the high-frequency components of the leakage current. Its execution flow is as follows: First, for The signal is bandpass filtered to extract components within the 2 kHz to 20 kHz frequency band. This bandpass filter employs a finite impulse response design with an order of 128 and exhibits linear phase characteristics.

[0041] Calculate the total effective value of this high-frequency component within the analysis time window, denoted as . This value reflects the overall intensity of partial discharge or corona activity on the insulator surface. Next, the cross-correlation functions of the high-frequency leakage current component I_hf and the amplitude sequences of the 3rd, 5th, and 7th harmonics of the voltage are calculated. The voltage harmonic amplitude sequences are obtained by performing a Fast Fourier Transform on the original voltage signal and extracting the amplitude at the corresponding frequency points. The calculation of the cross-correlation function is used to quantify the correlation strength between the high-frequency leakage current and the specific harmonic voltage in the time domain. As one embodiment of the invention, the calculation is implemented using a Fast Fourier Transform method with a Hanning window.

[0042] The specific process is: to The signal and target harmonic amplitude sequences are divided into multiple data segments, each with 1024 sampling points. Adjacent segments have a 50% overlap (512 sampling points). Each data segment is windowed and then subjected to a Fast Fourier Transform (FFT) to obtain its frequency domain representation. The cross-power spectrum is then calculated, and an inverse FFT is used to obtain the cross-correlation function sequence. The maximum peak value is identified from this sequence; this peak value is used as the characteristic quantity representing the modulation intensity of the leakage current by that harmonic. This process is performed on the 3rd, 5th, and 7th harmonics to obtain three cross-correlation peak characteristics, denoted as... .

[0043] The temperature compensation submodule receives real-time temperature data T from the ambient temperature acquisition unit. Its function is to eliminate the impact of ambient temperature changes on the conductivity of the insulating material (and consequently, the fundamental component of the leakage current), ensuring comparability of electrical characteristics under different climatic conditions. The theoretical basis for this is that the conductivity of the insulating material follows the Arrhenius model.

[0044] This submodule first extracts the effective value of the 50 Hz fundamental wave component from the raw leakage current signal before preprocessing using a phase-locked loop or synchronous sampling technology, denoted as . As one embodiment of the present invention, the Arrhenius model compensation function in the temperature compensation submodule is specifically expressed as: the effective value of the fundamental wave of the leakage current measured at the current temperature T degrees Celsius. The equivalent value at 25 degrees Celsius is calculated using the following formula. .

[0045] ;

[0046] in, The activation energy of an insulating material is expressed in electron volts. This is the Boltzmann constant, with a value of approximately 8.617333262145 × Electron volts per Kelvin; The current ambient temperature is expressed in degrees Celsius; in the formula... This converts Celsius to Kelvin; 298 is the Kelvin temperature corresponding to 25 degrees Celsius, which is 298 Kelvin. (Specific parameter values...) The calibration was obtained by performing temperature rise experiments on insulator samples of the same type. The calibration process was as follows: a constant voltage was applied to clean and dry insulator samples in a laboratory constant temperature chamber, and the fundamental component of the leakage current was measured at different temperatures. The results were then obtained by fitting the data. and The slope of the relationship curve is - Thus, to obtain The temperature compensation submodule internally stores calibrated values. The value is based on the real-time temperature T and the measured... It calculates and outputs the equivalent value of the fundamental frequency of the leakage current after temperature compensation in real time. .

[0047] The feature fusion submodule, as the final step in the multiphysics feature extraction module, is responsible for integrating the feature quantities generated by the aforementioned submodules into a unified feature vector. This submodule receives three cross-correlation peaks from the harmonic modulation analysis submodule. The equivalent fundamental value of leakage current after temperature compensation from the temperature compensation submodule. Furthermore, it also outputs directly from the signal preprocessing submodule. Two time-domain statistical features are extracted from the signal: pulse count and waveform kurtosis.

[0048] The method for extracting pulse counts is as follows: A dynamic threshold is set, and this threshold is set to... The signal amplitude is three times the effective value within the time window; the number of times the signal amplitude exceeds this dynamic threshold within the time window is counted as pulse count. Waveform kurtosis The calculation method is: Calculate The fourth central moment of the signal within the time window divided by the square of its variance, i.e. ,in Expressing expectations, This represents the mean. Kurtosis reflects the sharpness of the signal pulse impact. Finally, the feature fusion submodule combines these six feature quantities: , , , Combined into a 6-dimensional feature vector in a predetermined order. The feature vector is encapsulated into a data packet and sent to the adaptive fault diagnosis engine via the internal bus. Each feature vector data packet includes a timestamp of its generation and the corresponding ambient temperature value.

[0049] The adaptive fault diagnosis engine is the core decision-making unit of the system and is connected to the multiphysics feature extraction module. Please refer to the appendix. Figure 2The engine runs on the system's central processing unit in software, and its algorithmic logic includes a dynamic threshold update model and a fault state discriminator based on Mahalanobis distance. The engine maintains a health state benchmark dataset, which is stored in non-volatile memory.

[0050] A dynamic threshold update model is used to generate alarm thresholds for each feature in the feature vector that are adaptive to its historical operating status and ambient temperature. The model's establishment and operation are divided into two phases: an initialization phase and an online operation phase. The model is in the learning phase for the first 30 days after the system's initial commissioning.

[0051] During this period, the system continuously collects feature vector data of insulators under clear and dry weather conditions. After the adaptive fault diagnosis engine initially determines that there are no abnormalities, this data is marked as healthy state samples and stored in the healthy state benchmark dataset. After a 30-day learning period, statistical analysis is performed on all sample values ​​of each feature in the benchmark dataset. The arithmetic mean of each feature is calculated. and standard deviation The initial static threshold of this feature quantity. Set as For feature quantities requiring a lower limit alarm, the lower limit threshold can be set as follows: This yields the initial static thresholds for each of the six features.

[0052] Once the system enters normal operation, the dynamic threshold update model begins to function. This model automatically activates every 24 hours, and its task is to correct the initial static threshold and generate a dynamic prediction threshold based on the predicted ambient temperature for the next 24 hours. The correction process requires utilizing the relationship between feature quantities and temperature in historical data. As one embodiment of the present invention, the specific process of the linear regression correction is as follows: In the initialization phase, in addition to calculating the static threshold, the relationship between each feature quantity and historical ambient temperature is analyzed.

[0053] Specifically, using the ambient temperature data at the same time each day for the past 7 days as the independent variable and the measured value of the characteristic quantity at the corresponding time as the dependent variable, a straight line is fitted using the least squares method to obtain the slope k of the characteristic quantity's change with temperature. This slope reflects the degree of influence of temperature on the characteristic quantity. During system operation, it receives ambient temperature curves for the next 24 hours from weather forecasts or autoregressive predictions, and takes the average or highest value as the predicted temperature. Then, for this feature, the dynamic threshold for the next 24 hours... The calculation formula is: .in The baseline temperature is 25 degrees Celsius. Through offline training, a parameter table of temperature-feature relationship for different features and seasons can be generated. During online runtime, the slope is obtained by looking up the table. To improve computational efficiency, six dynamic thresholds are generated for each of the six features, forming a dynamic threshold vector. .

[0054] The fault state discriminator based on Mahalanobis distance is the core of the performance state classification. It receives the current feature vector from the multiphysics feature extraction module. And the dynamic threshold vectors of each feature provided by the dynamic threshold update model. The discrimination process is divided into two levels. The first level of discrimination is rapid screening. This level of discrimination will... The measured value of each feature quantity in Th is compared one by one with the corresponding dynamic threshold in Th.

[0055] The comparison logic is as follows: if the measured value of a certain characteristic exceeds its dynamic threshold upper limit or falls below its dynamic threshold lower limit, the characteristic is determined to be out of limit. The system is set to trigger conditions: if any two or more characteristic values ​​exceed their limits simultaneously, a Level 1 alarm is immediately triggered. A Level 1 alarm is considered a severe fault state, indicating that the insulator may have a serious risk of surface flashover or internal breakdown. This judgment result will skip subsequent discrimination and be sent to the output module. If only one characteristic value exceeds its limit or all characteristic values ​​do not exceed their limits, the system proceeds to Level 2 discrimination.

[0056] Level 2 discrimination is based on holistic pattern recognition using multivariate statistics. This discrimination calculates the current feature vector. The Mahalanobis distance relative to the health status benchmark dataset. The health status benchmark dataset, after initialization, includes not only the mean vector... Furthermore, the covariance matrix of its 6-dimensional eigenvectors was calculated. Mahalanobis distance The calculation formula is:

[0057] ;

[0058] in, Indicates transpose. This represents the inverse of the covariance matrix. Mahalanobis distance takes into account the correlation between features and can measure the degree to which the current state deviates from the overall health pattern. The system presets two distance thresholds: a warning threshold and a warning threshold. and alarm threshold .

[0059] The warning threshold is typically set as the value of a chi-square distribution with 6 degrees of freedom at a 95% confidence level, approximately the square root of 12.59. The alarm threshold is set as the value at a 99.9% confidence level, approximately the square root of 22.46. Specific thresholds can be fine-tuned using historical fault data. The logic for the second-level judgment is: if the calculated Mahalanobis distance... Greater than the warning threshold But less than the alarm threshold If the insulator is in an early abnormal state, a warning signal is generated. This state may correspond to latent faults such as the beginning of contamination accumulation or the appearance of micro-cracks inside, which have not yet caused a significant exceedance of any single characteristic value. If the Mahalanobis distance... Greater than or equal to alarm threshold If this condition is detected, it is determined to be a developing fault state, generating a level-two alarm signal, indicating that the abnormal mode has become very significant. If the Mahalanobis distance... Less than the warning threshold If the condition is met, it is determined to be in a normal state. All judgment results, including state labels, lists of features exceeding limits, calculated Mahalanobis distance values, and ratios of measured values ​​of each feature to the threshold, are encapsulated into intermediate diagnostic records.

[0060] The diagnostic result output and feedback module is connected to the adaptive fault diagnosis engine. Please refer to the appendix. Figure 5 This module is responsible for handling the final presentation and remote transmission of diagnostic results, as well as the closed-loop optimization of the system. First, it integrates the intermediate diagnostic records generated by the adaptive fault diagnosis engine with information such as ambient temperature, timestamps, and device identifiers, and encapsulates them into a diagnostic report conforming to a specific industry standard format.

[0061] The report format can use Extensible Markup Language (EXPLAIN) or JavaScript object representation, and includes a header, diagnostic body, checksum, etc. The diagnostic body details the diagnosis's conclusions, triggering criteria, characteristic data, and recommended measures. The packaged diagnostic report is uploaded to the remote monitoring master station through the communication interface integrated into this module. The communication interface supports multiple methods, including wired Ethernet, 4G or 5G wireless cellular networks, and power line carrier, and has automatic switching and retransmission mechanisms to ensure communication reliability.

[0062] At the same time, this module undertakes the crucial function of model feedback and updating. It will provide complete information from each diagnosis, including feature vectors. The final status label is then sent back to the adaptive fault diagnosis engine. For diagnostic records determined to be in a normal state or an early abnormal state, their corresponding feature vectors, after confirmation, are allowed to be used to update the health status benchmark dataset. This is to prevent fault data from contaminating the health benchmark. The system is designed with a rolling update algorithm to ensure that the diagnostic benchmark can adapt to the slow aging of the insulator material. As one embodiment of the present invention, the execution strategy of the rolling update algorithm is as follows: the system maintains a first-in-first-out rolling cache queue, the capacity of which corresponds to 90 days of normal status data. For each new diagnostic record determined to be in a normal state, its corresponding feature vector is added to the tail of the rolling cache queue, while the oldest record is removed from the head of the queue.

[0063] In this way, the queue always stores normal state feature samples from the most recent 90 days. When the system has been running for 90 days and the cache queue is full, the rolling update algorithm is officially launched. This algorithm does not update in real time, but is executed periodically at a low frequency, for example, once every 30 days. When an update is triggered, the algorithm uses all feature vector data in the current rolling cache queue to recalculate the statistical parameters of the health status baseline dataset.

Claims

1. A fault diagnosis system for an insulator detection device for measuring electrical variable, characterized in that, include: The synchronous data acquisition module is used to achieve high-precision time-synchronized data acquisition from the electrical variable measurement sensor group deployed on the target insulator string and environmental sensors; A multiphysics feature extraction module, connected to the synchronous data acquisition module, is used to preprocess the synchronously acquired raw data and extract feature vectors that can characterize the state of the insulator and are coupled with the influence of the electro-thermal environment. An adaptive fault diagnosis engine, connected to the multi-physics feature extraction module, is used to perform dynamic threshold comparison and state classification based on the input multi-dimensional feature vector; The diagnostic result output and feedback module is connected to the adaptive fault diagnosis engine. It is used to encapsulate the diagnostic status, specific out-of-limit features, Mahalanobis distance value and ambient temperature information into a standard format diagnostic report and upload it. At the same time, it sends back the feature vector and final status label of each diagnosis to the adaptive fault diagnosis engine. The multiphysics feature extraction module includes a harmonic modulation analysis submodule, a temperature compensation submodule, and a feature fusion submodule. The harmonic modulation analysis submodule is used to receive the preprocessed leakage current signal and voltage harmonic component data, calculate the total effective value of the leakage current signal in the 2 kHz to 20 kHz frequency band as the high-frequency leakage current component, and calculate the cross-correlation function between the high-frequency leakage current component and the amplitude of the 3rd, 5th and 7th harmonics of the voltage, respectively, and obtain the three cross-correlation peaks as characteristic quantities characterizing the modulation intensity of the harmonics on the leakage current. The temperature compensation submodule is used to receive ambient temperature data and construct a temperature compensation function based on the Arrhenius model of the conductivity of insulating materials. This function uses 25 degrees Celsius as a reference and converts the fundamental component of the leakage current at the current temperature to the equivalent value at the reference temperature in order to eliminate the influence of temperature on the conductivity of the insulating material. The feature fusion submodule is used to combine the three cross-correlation peak values ​​output by the harmonic modulation analysis submodule, the equivalent value of the fundamental frequency of the leakage current after temperature compensation output by the temperature compensation submodule, and the pulse count and waveform kurtosis extracted directly from the preprocessed leakage current signal, a total of six features, into a 6-dimensional feature vector and output it. The adaptive fault diagnosis engine incorporates a dynamic threshold update model and a fault state discriminator based on Mahalanobis distance. The dynamic threshold update model is used to generate an alarm threshold for each feature in the feature vector that is adapted to its historical operating state and ambient temperature; the fault state discriminator based on Mahalanobis distance is used to receive the current feature vector and the dynamic thresholds of each feature provided by the dynamic threshold update model, and perform two-level discrimination. The operation process of the dynamic threshold update model is as follows: During the first 30 days of system commissioning, feature vector data of insulators under sunny and dry weather is continuously collected and regarded as the health status benchmark dataset; For each feature in the benchmark dataset, calculate its mean and three times the standard deviation, and use them as the initial static threshold for that feature. Once the system is in normal operation, it performs linear regression correction on the initial static threshold every 24 hours based on the distribution of ambient temperature data over the past 7 days, generating a dynamic prediction threshold for the next 24 hours. The specific process of the linear regression correction is as follows: using ambient temperature as the independent variable and the average value of historical characteristic quantities as the dependent variable, the least squares method is used to fit and obtain the slope of temperature-characteristic quantity. The dynamic threshold under the future temperature prediction is equal to the initial static threshold plus the slope multiplied by the difference between the future temperature prediction and the base temperature of 25 degrees Celsius. The diagnostic result output and feedback module sends the feature vector and final status label of each diagnosis back to the health status benchmark dataset in the adaptive fault diagnosis engine, which is used to start the rolling update algorithm of the benchmark dataset after the system has run for 90 days. The execution strategy of the rolling update algorithm is as follows: the system maintains a first-in-first-out rolling cache queue with a capacity corresponding to 90 days of normal status data; for each new diagnostic record that is determined to be in a normal state, its corresponding feature vector is added to the tail of the rolling cache queue, and the oldest record is removed from the head of the queue; the rolling update algorithm uses all the feature vector data in the current rolling cache queue to recalculate the statistical parameters of the health status benchmark dataset.

2. The fault diagnosis system for an insulator detection device for measuring electrical variable as described in claim 1, characterized in that, The synchronous data acquisition module includes a voltage signal acquisition unit, a leakage current signal acquisition unit, and an ambient temperature acquisition unit. The voltage signal acquisition unit is used to acquire the power frequency voltage at the insulator operating point and its 2nd to 13th harmonic components, with a sampling frequency of not less than 10 kHz. The leakage current signal acquisition unit is used to synchronously acquire the full waveform leakage current signal of the insulator, and its measurement bandwidth covers 0 Hz to 1 MHz, with a dynamic range of not less than 120 dB. The ambient temperature acquisition unit is used to collect ambient temperature data at the insulator installation point in real time, with a measurement accuracy better than ±0.5 degrees Celsius. The three acquisition units are synchronized through a high-precision hardware clock source to ensure that all data samples have a unified timestamp and the time synchronization error is less than 1 microsecond.

3. The fault diagnosis system for an insulator detection device for measuring electrical variable as described in claim 1, characterized in that, The multiphysics feature extraction module also includes a signal preprocessing submodule; the signal preprocessing submodule is used to perform digital notch filtering on the original leakage current signal at 50 Hz power frequency and its integer multiples of harmonics to suppress system background interference, and to perform a 5-point linear moving average on the filtered signal to smooth random noise.

4. The fault diagnosis system for an insulator detection device for measuring electrical variable as described in claim 1, characterized in that, The two-level discrimination process of the fault state discriminator based on Mahalanobis distance is as follows: Level 1 discrimination, the measured value of each feature in the current feature vector is compared with the corresponding dynamic threshold one by one. If any two or more feature values ​​exceed the limit at the same time, a Level 1 alarm is immediately triggered and the fault state is judged to be serious. If only one feature exceeds the limit or all features do not exceed the limit, proceed to the second level of discrimination; The second level of discrimination calculates the Mahalanobis distance of the current feature vector relative to the health status benchmark dataset, and sets two distance thresholds: a warning threshold and an alarm threshold. If the Mahalanobis distance is greater than the warning threshold but less than the alarm threshold, it is determined to be an early abnormal state and a warning signal is generated. If the Mahalanobis distance is greater than or equal to the alarm threshold, it is determined to be a progressive fault state, and a level 2 alarm signal is generated; If the Mahalanobis distance is less than the warning threshold, it is determined to be in a normal state.

5. The fault diagnosis system for an insulator detection device for measuring electrical variable as described in claim 1, characterized in that, The cross-correlation function in the harmonic modulation analysis submodule is calculated using the Fast Fourier Transform method with a Hanning window, the window length is 1024 sampling points, and the overlap rate is 50%.

6. The fault diagnosis system for an insulator detection device for measuring electrical variable as described in claim 1 is characterized in that, The Arrhenius model compensation function in the temperature compensation submodule is specifically expressed as follows: The current temperature... The fundamental effective value of leakage current measured at Celsius Through formula Equivalent value at 25 degrees Celsius ;in The activation energy of insulating materials, This is the Boltzmann constant, and its specific value is obtained by calibrating it through temperature rise experiments on samples of the same type of insulator.

Images