Power transformation equipment insulation state evaluation and early warning method and system based on multi-sensor fusion

Abstract

The application discloses a power transformation equipment insulation state evaluation and early warning method and system based on multi-sensor fusion, and relates to the field of data processing.The method comprises the following steps: collecting synchronous data through distributed multi-sensor, carrying out layered denoising, abnormality processing, alignment completion and normalization preprocessing, extracting multi-source features such as acoustics, dielectric loss, leakage current, temperature and humidity, oil chromatography, combining weighted fusion and D-S evidence theory to generate a comprehensive feature set; quantitatively evaluating the insulation state based on a fuzzy comprehensive evaluation model and dividing the insulation state into four levels, and triggering corresponding early warning actions such as record, inspection, sound-light alarm and equipment isolation, and recording logs throughout the process; and continuously optimizing model parameters, feature extraction and early warning mechanism based on field feedback and historical data.The application has the advantages that: through multi-source information fusion and fuzzy comprehensive evaluation, accurate grading and grading early warning of the insulation state are realized, and at the same time, the reliability is continuously improved relying on the closed-loop optimization mechanism, thereby effectively guaranteeing the safe operation of the power transformation equipment.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method and system for assessing and warning the insulation status of substations using multi-sensor fusion. Background Technology

[0002] With the expansion of power grids, the increase in voltage levels, and the growing complexity of equipment, preventative testing and maintenance based on fixed-cycle outages are not only costly and inefficient, but also struggle to detect potential insulation degradation in equipment in real time. Modern power grids have increasingly stringent requirements for power supply reliability, and outage windows are shrinking, making condition-based maintenance an inevitable trend.

[0003] Most current methods for assessing and warning the insulation status of substation equipment rely on single-sensor monitoring or employ multi-sensor deployments lacking a systematic approach. They also fail to utilize time-scaled synchronization technology, leading to asynchronous and incomplete multi-dimensional data, and a tendency to create monitoring blind spots. Data preprocessing methods are simplistic, lacking tailored denoising and anomaly handling solutions for different data types, making it difficult to effectively eliminate interference and correct anomalies. Furthermore, they fail to adequately eliminate the influence of dimensions, resulting in insufficient data reliability. Simultaneously, feature extraction is redundant, and fusion methods are simplistic, failing to effectively resolve conflicts between multi-source information, leading to poor accuracy of comprehensive features. Status classification is ambiguous, and warning mechanisms are incomplete, lacking clear classification standards and standardized warning actions, and without full log recording, resulting in poor traceability. Moreover, most methods lack closed-loop optimization mechanisms, unable to adjust model parameters, feature extraction strategies, and sensor deployment based on on-site verification feedback. Long-term use is prone to false and missed warnings, making it difficult to adapt to the complex and ever-changing operating environment of substation equipment, and hindering the continuous improvement of the accuracy and reliability of assessment and warning systems. Summary of the Invention

[0004] To improve existing methods and systems, this paper presents a method and system for assessing and warning the insulation status of substation equipment using multi-sensor fusion. This method achieves accurate grading and early warning of insulation status through hierarchical preprocessing, multi-source information fusion, and fuzzy comprehensive evaluation. It also continuously improves reliability through a closed-loop optimization mechanism, effectively ensuring the safe operation of substation equipment.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A multi-sensor fusion-based method for assessing and warning the insulation status of substation equipment includes:

[0007] Multiple sensors are distributed and deployed in the insulation parts of power equipment. Time synchronization technology is used to unify the time of multi-source data and complete the real-time acquisition of multi-dimensional insulation data.

[0008] The collected raw data is processed in layers, and corresponding denoising methods are used for different types of data. Outliers are identified and classified. Data alignment and missing value supplementation are completed based on timestamps. The influence of units is eliminated through normalization.

[0009] Based on the preprocessed multi-source data, the time domain and phase characteristics of the ultrasonic signal, the relevant parameters of dielectric loss and leakage current, the temperature and humidity difference and rate of change, and the concentration and ratio of characteristic gases in the insulating oil are extracted. After removing redundant features, a single sensor insulation feature set is formed.

[0010] Combining weighted fusion and DS evidence theory, weights are assigned to the feature sets of each single sensor. Preliminary fusion features are obtained by weighted summation. Information redundancy and conflict are eliminated by evidence synthesis. After normalization, a comprehensive insulation feature set is obtained.

[0011] A four-level insulation status classification standard is established. Combining comprehensive feature set, operating standards and historical data, a fuzzy comprehensive evaluation model is adopted. The quantitative evaluation is carried out through membership function and weight calculation to obtain the current insulation level of the equipment.

[0012] Based on the four-level insulation status, four levels of early warning are set. The first level of early warning only records the status. The second level triggers a yellow early warning and is included in the inspection. The third level triggers an orange audible and visual alarm and pushes information. The fourth level triggers a red early warning, isolates the equipment and pushes an emergency notification. All early warning actions are logged in the entire process.

[0013] Collect on-site verification and handling data to determine the accuracy of early warnings. Optimize the evaluation model parameters and feature extraction process for false and missed early warnings. Optimize sensor deployment and early warning mechanisms by regularly analyzing historical data.

[0014] Preferably, the distributed deployment of multiple sensors on the insulation parts of the power equipment, and the use of time-stamping synchronization technology to unify the time of multi-source data and complete the real-time acquisition of multi-dimensional insulation data, specifically includes:

[0015] Various types of sensors are distributed and deployed on the insulation parts of power equipment. These sensors include ultrasonic sensors, dielectric loss testers, Hall leakage current sensors, temperature sensors, humidity sensors, and gas chromatographs.

[0016] When any data point is detected to be close to a preset threshold, the collection frequency will be increased, and a unified timestamp will be added to all collected data through time stamp synchronization technology.

[0017] Preferably, the layered processing of the collected raw data, employing corresponding denoising methods for different types of data, identifying and classifying outliers, completing data alignment and missing value supplementation based on timestamps, and eliminating the influence of dimensions through normalization processing specifically includes:

[0018] For different types of data, corresponding noise reduction methods are adopted. Ultrasonic partial discharge signals are decomposed into three levels using the db4 wavelet basis to remove high-frequency interference noise. Dielectric loss data and leakage current signals are processed using the moving average method, replacing the current data with the average value of data from five consecutive acquisition cycles. Characteristic gas concentration data are processed using an adaptive filtering method to filter out concentration abrupt changes.

[0019] The 3σ criterion is used to identify abnormal data. For occasional abnormal values, linear interpolation of two adjacent valid data is used for replacement. For abnormal values ​​that occur in three or more consecutive periods, they are determined to be sensor faults.

[0020] Based on the added unified timestamp, data from the same time point collected by different sensors are matched, and linear interpolation is used to supplement the missing small amount of data.

[0021] The min-max normalization method is used to map all preprocessed data to the [0,1] interval to eliminate different dimensions.

[0022] Preferably, the step of extracting the time-domain and phase characteristics of the ultrasonic signal, relevant parameters of dielectric loss and leakage current, temperature and humidity difference and rate of change, and characteristic gas concentration and ratio of insulating oil based on preprocessed multi-source data, and then removing redundant features to form a single-sensor insulation feature set specifically includes:

[0023] For ultrasonic partial discharge data, time-domain features and phase features are extracted. For dielectric loss data, real-time dielectric loss value, rate of change of dielectric loss value, and amplitude of dielectric loss value fluctuation are extracted. For leakage current data, peak current value, effective current value, harmonic components of current, and current change trend are extracted. For temperature and humidity data, ambient humidity value, equipment surface temperature value, temperature difference value, and rate of change of temperature and humidity are extracted. For characteristic gas data of insulating oil, concentration values ​​of methane, ethane, ethylene, and acetylene and the rate of change of each gas concentration are extracted, and gas concentration ratio parameters are also extracted.

[0024] All extracted feature parameters were validated for effectiveness, and redundant features that were not related to the insulation state were removed to form a single-sensor insulation feature set.

[0025] Preferably, the step of combining weighted fusion and DS evidence theory to assign weights to the feature sets of each single sensor, obtaining preliminary fused features through weighted summation, eliminating information redundancy and conflict through evidence synthesis, and obtaining a comprehensive insulation feature set after normalization specifically includes:

[0026] Based on the measurement accuracy, data reliability, and correlation between characteristic parameters and insulation status of each sensor, the weights of the feature sets of each individual sensor are obtained.

[0027] The corresponding feature parameters in each individual sensor feature set are weighted and summed according to the assigned weights to obtain preliminary fused features;

[0028] The DS evidence theory is used to perform secondary fusion of the initial fusion features, treating each initial fusion feature as independent evidence, determining the basic probability allocation function of each piece of evidence, and synthesizing all the evidence through evidence synthesis rules;

[0029] The fused feature set is normalized to obtain the final integrated insulation feature set.

[0030] Preferably, the establishment of a four-level insulation status grading standard, combined with a comprehensive feature set, operating standards, and historical data, employs a fuzzy comprehensive evaluation model, and performs quantitative evaluation through membership functions and weight calculations to obtain the current insulation level of the equipment. Specifically, this includes:

[0031] The insulation condition is divided into four levels: normal condition, warning condition, abnormal condition, and critical condition.

[0032] A delay determination mechanism is set up so that when the comprehensive feature parameter is continuously lower or higher than the corresponding threshold for more than 60 seconds, it is determined to be the corresponding state.

[0033] A fuzzy comprehensive evaluation model is used for hierarchical evaluation. The comprehensive insulation feature set is used as the evaluation factor set, and the four-level insulation status standard is used as the evaluation comment set. The membership function of each evaluation factor is calculated.

[0034] The weights of each evaluation factor are determined by the analytic hierarchy process (AHP), and fuzzy synthesis is performed to obtain a comprehensive evaluation result of the equipment insulation status and output the current insulation status level of the equipment.

[0035] Preferably, the four-level early warning system is set based on the four-level insulation status. Level 1 warning only records the status; Level 2 triggers a yellow warning and is included in the inspection; Level 3 triggers an orange audible and visual alarm and pushes information; Level 4 triggers a red warning, isolates the equipment, and sends an emergency notification. All warning actions are logged in their entirety, including:

[0036] Corresponding to the four-level insulation status standard, four levels of early warning are set. Level 1 early warning corresponds to the normal state and does not trigger any early warning action; Level 2 early warning corresponds to the attention state, triggers a yellow warning, records the early warning log, and incorporates the early warning information into the inspection; Level 3 early warning corresponds to the abnormal state, triggers an orange audible and visual alarm, points out the faulty equipment and abnormal characteristic parameters; Level 4 early warning corresponds to the critical state, triggers the highest level red audible and visual alarm, physically isolates the faulty equipment through the linkage equipment protection system, and pushes the fault handling plan.

[0037] All warning actions are logged, including the warning trigger time, warning level, abnormal parameters, and executed actions.

[0038] Preferably, the process of collecting on-site verification and handling data, judging the accuracy of early warnings, optimizing the evaluation model parameters and feature extraction steps for false and missed early warnings, and optimizing sensor deployment and early warning mechanisms through regular analysis of historical data specifically includes:

[0039] After on-site verification, the verification results are entered into the data processing terminal. By sorting and analyzing the feedback data, the accuracy of the warning trigger and the rationality of the warning level are judged.

[0040] For false alarms, analyze the causes of false alarms, adjust the evaluation threshold, membership function or feature weights, and optimize the evaluation model.

[0041] For leakage warnings, supplement fault data, add relevant characteristic parameters, regularly conduct statistical analysis on historical warning data and fault data, generate insulation fault analysis reports, and optimize sensor deployment locations, acquisition frequencies, and warning mechanisms.

[0042] Furthermore, a multi-sensor fusion-based substation insulation condition assessment and early warning system is proposed, including:

[0043] Data acquisition and preprocessing module: Distributed deployment of multiple sensors to perform layered processing on raw data, including noise reduction, outlier identification and processing, data alignment and missing value filling, and normalization;

[0044] Feature extraction and filtering module: Extracts various insulation feature parameters from the preprocessed data, removes redundant features, and forms a single sensor feature set;

[0045] Multi-source information fusion module: Combining weighted fusion and DS evidence theory, it performs weight allocation, preliminary fusion and evidence synthesis on single sensor features to generate a comprehensive insulation feature set;

[0046] Condition assessment and classification module: Based on the fuzzy comprehensive evaluation model, combined with operating standards and historical data, calculates the membership degree and level of insulation condition;

[0047] Early warning and logging module: Triggers corresponding early warnings based on insulation level, performs actions such as recording, inspection notification, audible and visual alarms or equipment isolation, and logs the entire process.

[0048] Model optimization and feedback module: collects on-site verification data, analyzes the accuracy of early warnings, and optimizes and evaluates model parameters, feature extraction strategies, and sensor deployment;

[0049] Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

[0050] Compared with the prior art, the advantages of the present invention are:

[0051] This system achieves comprehensive, accurate, and closed-loop optimization of insulation condition assessment. By distributing multiple sensors and combining them with time-stamped synchronization technology, it enables real-time synchronous acquisition of multi-dimensional insulation data, compensating for the blind spots of single-sensor monitoring. Layered preprocessing employs adaptive denoising and anomaly handling schemes for different data characteristics, combined with normalization to eliminate the influence of dimensions, ensuring data reliability. Based on multi-source feature extraction and redundancy removal, combined with weighted fusion and DS evidence theory, it effectively solves information redundancy and conflicts, improving the accuracy of the comprehensive feature set. A four-level state classification and early warning mechanism accurately matches the insulation status of equipment, with standardized early warning actions and full traceability, balancing safety and traceability. Simultaneously, on-site verification feedback optimizes model parameters, feature extraction, and sensor deployment, forming a closed loop of "acquisition-processing-assessment-early warning-optimization," significantly improving the scientific rigor of substation insulation condition assessment and the timeliness of early warnings, reducing the risk of equipment insulation failures, and ensuring the safe and stable operation of the power grid. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the method proposed in this invention;

[0053] Figure 2 This is a schematic diagram of multi-source insulation data acquisition proposed in this invention;

[0054] Figure 3 This is a schematic diagram of the data preprocessing proposed in this invention;

[0055] Figure 4 This is a schematic diagram of the insulation state feature extraction proposed in this invention;

[0056] Figure 5 This is a schematic diagram of the multi-sensor feature-level data fusion proposed in this invention;

[0057] Figure 6 This is a schematic diagram of the insulation state classification assessment proposed in this invention;

[0058] Figure 7 This is a schematic diagram illustrating the hierarchical early warning triggering and execution proposed in this invention;

[0059] Figure 8 This is a schematic diagram of the early warning feedback and optimization proposed in this invention. Detailed Implementation

[0060] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0061] A multi-sensor fusion-based substation insulation condition assessment and early warning system includes:

[0062] Data acquisition and preprocessing module: Distributed deployment of multiple sensors to perform layered processing on raw data, including noise reduction, outlier identification and processing, data alignment and missing value filling, and normalization;

[0063] Feature extraction and filtering module: Extracts various insulation feature parameters from the preprocessed data, removes redundant features, and forms a single sensor feature set;

[0064] Multi-source information fusion module: Combining weighted fusion and DS evidence theory, it performs weight allocation, preliminary fusion and evidence synthesis on single sensor features to generate a comprehensive insulation feature set;

[0065] Condition assessment and classification module: Based on the fuzzy comprehensive evaluation model, combined with operating standards and historical data, calculates the membership degree and level of insulation condition;

[0066] Early warning and logging module: Triggers corresponding early warnings based on insulation level, performs actions such as recording, inspection notification, audible and visual alarms or equipment isolation, and logs the entire process.

[0067] Model optimization and feedback module: collects on-site verification data, analyzes the accuracy of early warnings, and optimizes and evaluates model parameters, feature extraction strategies, and sensor deployment;

[0068] Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

[0069] See Figure 1 As shown, the multi-sensor fusion method for assessing and warning the insulation status of substation equipment includes:

[0070] Step 1: Distribute multiple sensors to the insulation parts of the power equipment, and use time-stamping synchronization technology to unify the time of multi-source data to complete the real-time acquisition of multi-dimensional insulation data;

[0071] Step 2: The collected raw data is processed in layers. Corresponding denoising methods are used for different types of data. Outliers are identified and classified. Data alignment and missing value supplementation are completed based on timestamps. The influence of units is eliminated through normalization.

[0072] Step 3: Based on the preprocessed multi-source data, extract the time domain and phase characteristics of the ultrasonic signal, the relevant parameters of dielectric loss and leakage current, the temperature and humidity difference and rate of change, and the concentration and ratio of characteristic gases in the insulating oil. After removing redundant features, a single sensor insulation feature set is formed.

[0073] Step 4: Combining weighted fusion and DS evidence theory, weights are assigned to the feature sets of each single sensor. Preliminary fusion features are obtained through weighted summation. Information redundancy and conflict are eliminated through evidence synthesis. After normalization, a comprehensive insulation feature set is obtained.

[0074] Step 5: Establish a four-level insulation status classification standard. Combining the comprehensive feature set, operating standards and historical data, adopt a fuzzy comprehensive evaluation model, and perform quantitative evaluation through membership function and weight calculation to obtain the current insulation level of the equipment.

[0075] Step Six: Set up four levels of early warning based on the four-level insulation status. Level 1 early warning only records the status, Level 2 triggers a yellow early warning and is included in the inspection, Level 3 triggers an orange audible and visual alarm and pushes information, and Level 4 triggers a red early warning, isolates the equipment and pushes an emergency notification. All early warning actions are logged throughout the process.

[0076] Step 7: Collect on-site verification and handling data, determine the accuracy of the early warning, optimize the evaluation model parameters and feature extraction process for false and missed early warnings, and optimize sensor deployment and early warning mechanism by regularly analyzing historical data.

[0077] See Figure 2 As shown, multiple sensors are distributed and deployed on the insulation parts of power equipment. Time synchronization technology is used to unify the time of multi-source data, enabling real-time acquisition of multi-dimensional insulation data. Specifically, this includes:

[0078] Various types of sensors are distributed and deployed on the insulation parts of power equipment. These sensors include ultrasonic sensors, dielectric loss testers, Hall leakage current sensors, temperature sensors, humidity sensors, and gas chromatographs.

[0079] When any data point is detected to be close to a preset threshold, the collection frequency will be increased, and a unified timestamp will be added to all collected data through time stamp synchronization technology.

[0080] See Figure 3 As shown, the collected raw data is processed in layers, with corresponding denoising methods used for different data types. Outliers are identified and classified, data alignment and missing value imputation are performed based on timestamps, and the influence of units is eliminated through normalization. Specifically, this includes:

[0081] For different types of data, corresponding noise reduction methods are adopted. Ultrasonic partial discharge signals are decomposed into three levels using the db4 wavelet basis to remove high-frequency interference noise. Dielectric loss data and leakage current signals are processed using the moving average method, replacing the current data with the average value of data from five consecutive acquisition cycles. Characteristic gas concentration data are processed using an adaptive filtering method to filter out concentration abrupt changes.

[0082] The 3σ criterion is used to identify abnormal data. For occasional abnormal values, linear interpolation of two adjacent valid data is used for replacement. For abnormal values ​​that occur in three or more consecutive periods, they are determined to be sensor faults.

[0083] Based on the added unified timestamp, data from different sensors at the same time point are matched, and linear interpolation is used to supplement the missing small amount of data.

[0084] The min-max normalization method is used to map all preprocessed data to the [0,1] interval to eliminate different dimensions.

[0085] Specifically, a dedicated denoising scheme is adopted based on the noise characteristics of different types of data. For ultrasonic partial discharge signals, which are easily affected by equipment vibration and environmental noise, a 3-level wavelet decomposition using a dB4 wavelet basis is employed. This decomposes the signal into high-frequency noise components and low-frequency effective components. By setting a reasonable threshold, wavelet coefficients corresponding to high-frequency noise are removed. Then, wavelet reconstruction is performed on the remaining low-frequency components to restore a pure partial discharge ultrasonic signal. For dielectric loss data and leakage current signals, which mainly exhibit instantaneous fluctuation noise, a 5-point moving average method is used. Centered on the current acquisition period data, effective data from two adjacent periods are selected, and the arithmetic mean of the five data points is calculated. This average is then used to replace the current acquisition data, effectively eliminating the deviation caused by instantaneous fluctuations. For characteristic gas concentration data, which are susceptible to sudden concentration changes due to environmental airflow and sampling pipeline interference, an adaptive filtering method is used. This method captures data change trends in real time, automatically adjusts the filtering coefficients, filters out sudden abnormal fluctuations, and preserves the true variation pattern of gas concentration.

[0086] The accurate handling of outliers employs the 3σ criterion for outlier identification and correction. First, statistical analysis is performed on the raw data of each type to calculate the historical mean and standard deviation of that type of data. Data that deviates from the mean by more than three times the standard deviation are identified as outliers. For occasional single outliers, linear interpolation is used for correction. Two adjacent valid data points before and after the outlier are selected, and the correction value corresponding to the outlier is calculated based on the difference between the two data points and the time interval. For outliers that occur in three or more consecutive acquisition cycles, the corresponding sensor is identified as faulty, and the sensor number, fault type, and outlier time period are immediately marked on the data processing terminal.

[0087] See Figure 4 As shown, based on the preprocessed multi-source data, the time-domain and phase characteristics of the ultrasonic signal, relevant parameters of dielectric loss and leakage current, temperature and humidity difference and rate of change, and characteristic gas concentration and ratio of insulating oil are extracted. After removing redundant features, a single-sensor insulation feature set is formed, which specifically includes:

[0088] For ultrasonic partial discharge data, time-domain features and phase features are extracted. For dielectric loss data, real-time dielectric loss value, rate of change of dielectric loss value, and amplitude of dielectric loss value fluctuation are extracted. For leakage current data, peak current value, effective current value, harmonic components of current, and current change trend are extracted. For temperature and humidity data, ambient humidity value, equipment surface temperature value, temperature difference value, and rate of change of temperature and humidity are extracted. For characteristic gas data of insulating oil, concentration values ​​of methane, ethane, ethylene, and acetylene and the rate of change of each gas concentration are extracted, and gas concentration ratio parameters are also extracted.

[0089] All extracted feature parameters were validated for effectiveness, and redundant features that were not related to the insulation state were removed to form a single-sensor insulation feature set.

[0090] Specifically, for ultrasonic partial discharge preprocessing data, a combination of time-domain analysis and phase analysis is used to extract features. In time-domain feature extraction, a 10-minute analysis cycle is set. Valid discharge signals are determined by the signal amplitude threshold. The number of signal peaks exceeding the threshold within the cycle is counted, which is the number of discharges. The maximum value among all valid peaks is selected as the discharge amplitude peak value, and the arithmetic mean of all valid peaks is calculated as the discharge amplitude average value. The time interval from when the amplitude of each valid discharge signal reaches the threshold to when it falls below the threshold is recorded sequentially, and the arithmetic mean of all intervals is taken as the discharge duration.

[0091] Phase feature extraction converts the ultrasonic signal into a phase spectrum using Fourier transform, analyzes the signal energy distribution in the phase spectrum, and determines the phase range covered by the discharge signal. The percentage of discharges in each phase interval is statistically analyzed, and continuous phase intervals with a percentage exceeding 50% are identified as phase concentration intervals.

[0092] For dielectric loss preprocessing data, three core features are extracted: the real-time dielectric loss value is directly taken from the preprocessed current period data; the dielectric loss value change rate is calculated by dividing the difference between the dielectric loss value of the current period and the previous period by the time interval between the two periods.

[0093] The fluctuation range of dielectric loss is statistically analyzed using a 1-hour period. The difference between the maximum and minimum values ​​of dielectric loss within this period is calculated to reflect the stability of the dielectric loss. For leakage current preprocessing data, time-domain analysis is used to extract the peak current and the effective current value.

[0094] Frequency domain analysis uses Fast Fourier Transform to separate the 3rd and 5th harmonic components, extract the amplitude of each harmonic component, and perform linear fitting on the effective current value of 10 consecutive acquisition cycles. The upward or downward trend of the current is determined by the slope of the fitted line. A positive slope indicates an upward trend, while a negative slope indicates a downward trend.

[0095] For temperature and humidity preprocessing data, the real-time values ​​of ambient humidity and equipment surface temperature are directly extracted; the difference between equipment surface temperature and ambient temperature is calculated to eliminate the interference of ambient temperature on equipment temperature; the rate of change of temperature and humidity is calculated by dividing the difference between the values ​​of the current period and the previous period by the time interval.

[0096] For the preprocessing data of characteristic gases in insulating oil, the real-time concentration values ​​of methane, ethane, ethylene, and acetylene, as well as the concentration change rate of each gas, are extracted. At the same time, the concentration ratios of acetylene to ethylene and methane to ethane are calculated to help determine the type of insulation degradation.

[0097] See Figure 5 As shown, combining weighted fusion and DS evidence theory, weights are assigned to the feature sets of each single sensor. Preliminary fused features are obtained through weighted summation. Information redundancy and conflict are eliminated through evidence synthesis. After normalization, the comprehensive insulation feature set is obtained, specifically including:

[0098] Based on the measurement accuracy, data reliability, and correlation between characteristic parameters and insulation status of each sensor, the weights of the feature sets of each individual sensor are obtained.

[0099] The corresponding feature parameters in each individual sensor feature set are weighted and summed according to the assigned weights to obtain preliminary fused features;

[0100] The DS evidence theory is used to perform secondary fusion of the initial fusion features, treating each initial fusion feature as independent evidence, determining the basic probability allocation function of each piece of evidence, and synthesizing all the evidence through evidence synthesis rules;

[0101] The fused feature set is normalized to obtain the final integrated insulation feature set.

[0102] Specifically, the feature weights are scientifically allocated using Pearson correlation analysis to calculate the correlation coefficient between each individual sensor feature set and the insulation fault state of the substation equipment. Weights are initially assigned based on the measurement accuracy of each sensor. Partial discharge, dielectric loss, and leakage current features are most strongly correlated with insulation degradation, and their weights are set to 0.3, 0.25, and 0.2, respectively. Temperature and humidity features, and the gas ratio of insulating oil features are weighted at 0.15 and 0.1, respectively. Further optimization is achieved through historical fault data. One hundred sets of fault and normal data from similar equipment over the past three years are selected to compare the fusion effects under different weight combinations, and the weight values ​​are adjusted accordingly.

[0103] Next, a weighted fusion process is performed. Similar feature parameters from each individual sensor feature set are precisely matched, and the same type of feature parameters are summed according to assigned weights to obtain a preliminary fused feature set. Then, a secondary fusion is performed using DS evidence theory. Each feature in the preliminary fused feature set is treated as independent evidence. Based on the magnitude of each feature parameter, a basic probability allocation function is determined for each piece of evidence—the closer a feature parameter is to the fault threshold, the higher the basic probability value of the corresponding evidence, and vice versa. Using DS evidence synthesis rules, all independent evidence is synthesized, and the conflict coefficient between each piece of evidence is calculated. If the conflict coefficient is higher than 0.8, a weighted average method is used to correct the evidence and reduce the impact of conflict. If the conflict coefficient is lower than 0.8, evidence synthesis is performed directly to obtain a comprehensive probability allocation result. The feature combination with the highest probability is selected to form the secondary fused feature set.

[0104] See Figure 6 As shown, a four-level insulation status classification standard is established. Combining a comprehensive feature set, operating standards, and historical data, a fuzzy comprehensive evaluation model is adopted. Quantitative evaluation is performed through membership functions and weight calculations to obtain the specific current insulation level of the equipment, including:

[0105] The insulation condition is divided into four levels: normal condition, warning condition, abnormal condition, and critical condition.

[0106] A delay determination mechanism is set up so that when the comprehensive feature parameter is continuously lower or higher than the corresponding threshold for more than 60 seconds, it is determined to be the corresponding state.

[0107] A fuzzy comprehensive evaluation model is used for hierarchical evaluation. The comprehensive insulation feature set is used as the evaluation factor set, and the four-level insulation status standard is used as the evaluation comment set. The membership function of each evaluation factor is calculated.

[0108] The weights of each evaluation factor are determined by the analytic hierarchy process (AHP), and fuzzy synthesis is performed to obtain a comprehensive evaluation result of the equipment insulation status and output the current insulation status level of the equipment.

[0109] Specifically, a fuzzy comprehensive evaluation model is used to achieve quantitative grading. The resulting comprehensive insulation feature set is used as the evaluation factor set, covering all core integrated features such as discharge characteristics, dielectric loss characteristics, and leakage current characteristics. The four levels of status—normal, warning, abnormal, and severe—are used as the evaluation comment set. The membership degree of each evaluation factor is calculated using a trapezoidal membership function. Based on the range of each feature parameter, the membership degree value (range 0-1) corresponding to each insulation level is determined. The closer the membership degree is to 1, the more the feature is inclined to the corresponding status. The weight of each evaluation factor is determined by the analytic hierarchy process (AHP), and a judgment matrix is ​​constructed. The importance of each feature is compared pairwise, and the consistency test ensures that the weight allocation is reasonable. The partial discharge fusion feature and leakage current fusion feature have the highest weights, accounting for 0.3 and 0.25 respectively, the dielectric loss fusion feature accounts for 0.2, and the other features account for 0.25. The weighted average method is used for fuzzy synthesis calculation. The membership degree of each evaluation factor is multiplied by its corresponding weight and then summed to obtain the comprehensive membership degree of the equipment corresponding to each insulation level. The status with the highest comprehensive membership degree is selected as the current insulation status level of the equipment.

[0110] The formula for fuzzy synthesis is:

[0111]

[0112] in, This is a vector representing the comprehensive assessment result of the equipment insulation status. The evaluation factor weight vector represents the weight of each comprehensive insulation characteristic in the assessment. , For fuzzy synthesis operators, This is a membership matrix, where each row corresponds to an insulation feature and each column corresponds to an insulation state, storing the membership degree of each feature to each state level. Let i be the weight value of the i-th integrated insulation feature. Let be the membership degree of the i-th feature to the k-th insulation state.

[0113] See Figure 7 As shown, a four-level early warning system is set based on the four-level insulation status. The first-level early warning only records the status; the second-level system triggers a yellow early warning and includes it in the inspection; the third-level system triggers an orange audible and visual alarm and pushes information; and the fourth-level system triggers a red early warning, isolates the equipment, and sends an emergency notification. All early warning actions are logged in the entire process, including:

[0114] Corresponding to the four-level insulation status standard, four levels of early warning are set. Level 1 early warning corresponds to the normal state and does not trigger any early warning action; Level 2 early warning corresponds to the attention state, triggers a yellow warning, records the early warning log, and incorporates the early warning information into the inspection; Level 3 early warning corresponds to the abnormal state, triggers an orange audible and visual alarm, points out the faulty equipment and abnormal characteristic parameters; Level 4 early warning corresponds to the critical state, triggers the highest level red audible and visual alarm, physically isolates the faulty equipment through the linkage equipment protection system, and pushes the fault handling plan.

[0115] All warning actions are logged, including the warning trigger time, warning level, abnormal parameters, and executed actions.

[0116] Specifically, Level 1 warning corresponds to the normal state, and the trigger condition is that all comprehensive insulation characteristic parameters are within the normal threshold range. The only action is that the system automatically records the current insulation status of the equipment, the values ​​of each characteristic parameter, and the timestamp. Level 2 warning corresponds to the attention state, and the trigger condition is that the comprehensive evaluation result is in the attention state, triggering a yellow warning, recording the warning log, not issuing audible or visual alarms, and only including the equipment warning information in the routine inspection plan.

[0117] A Level 3 warning indicates an abnormal state. If three or more parameters are within the warning range or one or two parameters are within the abnormal range, an orange warning is triggered. This automatically initiates correlation analysis, simultaneously retrieving the equipment's insulation data for the past 72 hours, recent operation records, and operating parameters of related equipment to generate a trend curve.

[0118] Level 4 warning indicates a severe situation. Three or more parameters are in abnormal ranges or any parameter is in a dangerous range, triggering the highest red warning. Immediately activate red audible and visual alarms and a cyclical voice alarm, and automatically execute protective controls: lock the automatic reclosing function of the equipment, disconnect the upstream switch to achieve physical isolation, and push a detailed emergency response plan to the terminal.

[0119] See Figure 8 As shown, data from on-site verification and handling are collected to determine the accuracy of early warnings. The evaluation model parameters and feature extraction process are optimized to address false and missed warnings. Specifically, sensor deployment and early warning mechanisms are optimized through regular analysis of historical data.

[0120] After on-site verification, the verification results are entered into the data processing terminal. By sorting and analyzing the feedback data, the accuracy of the warning trigger and the rationality of the warning level are judged.

[0121] For false alarms, analyze the causes of false alarms, adjust the evaluation threshold, membership function or feature weights, and optimize the evaluation model.

[0122] For leakage warnings, supplement fault data, add relevant characteristic parameters, regularly conduct statistical analysis on historical warning data and fault data, generate insulation fault analysis reports, and optimize sensor deployment locations, acquisition frequencies, and warning mechanisms.

[0123] Specifically, the feedback data is divided into three categories: false alarms, missed alarms, and accurate alarms. The criteria for judging false alarms is that the deviation between the system's assessed state and the actual state on site is one level or more. The criteria for judging missed alarms is that the equipment is confirmed to have an insulation fault on site, but the system does not trigger the corresponding level of alarm.

[0124] For false alarms, analyze the causes one by one. If it is caused by an unreasonable assessment threshold, adjust the threshold range of the corresponding feature parameters by combining historical normal data and on-site verification results, and optimize the interval division of the membership function. If it is caused by a deviation in feature weight allocation, recalculate the correlation between each feature and insulation fault, and adjust the weight values ​​to ensure that the weight matches the importance of the feature.

[0125] In response to missed warnings, the corresponding fault data will be added to the model training database to improve the feature extraction process. If the missed warning is caused by the failure to extract key features, relevant feature parameters will be added, the feature extraction rules will be re-optimized, redundant features will be removed, and effective features will be retained. At the same time, the rationality of sensor deployment will be checked. If the missed data collection is caused by insufficient sensor coverage, the sensor deployment location will be adjusted or the number of sensors will be increased to ensure comprehensive collection of multi-source data.

[0126] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0127] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing and warning the insulation status of substation equipment using multi-sensor fusion, characterized in that, include: Multiple sensors are distributed and deployed in the insulation parts of power equipment. Time synchronization technology is used to unify the time of multi-source data and complete the real-time acquisition of multi-dimensional insulation data. The collected raw data is processed in layers, and corresponding denoising methods are used for different types of data. Outliers are identified and classified. Data alignment and missing value supplementation are completed based on timestamps. The influence of units is eliminated through normalization. Based on the preprocessed multi-source data, the time domain and phase characteristics of the ultrasonic signal, the relevant parameters of dielectric loss and leakage current, the temperature and humidity difference and rate of change, and the concentration and ratio of characteristic gases in the insulating oil are extracted. After removing redundant features, a single sensor insulation feature set is formed. Combining weighted fusion and DS evidence theory, weights are assigned to the feature sets of each single sensor. Preliminary fusion features are obtained by weighted summation. Information redundancy and conflict are eliminated by evidence synthesis. After normalization, a comprehensive insulation feature set is obtained. A four-level insulation status classification standard is established. Combining comprehensive feature set, operating standards and historical data, a fuzzy comprehensive evaluation model is adopted. The quantitative evaluation is carried out through membership function and weight calculation to obtain the current insulation level of the equipment. Based on the four-level insulation status, four levels of early warning are set. The first level of early warning only records the status. The second level triggers a yellow early warning and is included in the inspection. The third level triggers an orange audible and visual alarm and pushes information. The fourth level triggers a red early warning, isolates the equipment and pushes an emergency notification. All early warning actions are logged in the entire process. Collect on-site verification and handling data to determine the accuracy of early warnings. Optimize the evaluation model parameters and feature extraction process for false and missed early warnings. Optimize sensor deployment and early warning mechanisms by regularly analyzing historical data.

2. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The specific steps involve distributing multiple sensors to the insulation components of power equipment and using time-stamping synchronization technology to unify the time of multi-source data, thereby achieving real-time acquisition of multi-dimensional insulation data. Various types of sensors are distributed and deployed on the insulation parts of power equipment. These sensors include ultrasonic sensors, dielectric loss testers, Hall leakage current sensors, temperature sensors, humidity sensors, and gas chromatographs. When any data point is detected to be close to a preset threshold, the collection frequency will be increased, and a unified timestamp will be added to all collected data through time stamp synchronization technology.

3. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The process of stratifying the collected raw data, employing corresponding denoising methods for different data types, identifying and classifying outliers, aligning data and filling in missing values ​​based on timestamps, and eliminating the influence of units through normalization specifically includes: For different types of data, corresponding noise reduction methods are adopted. Ultrasonic partial discharge signals are decomposed into three levels using the db4 wavelet basis to remove high-frequency interference noise. Dielectric loss data and leakage current signals are processed using the moving average method, replacing the current data with the average value of data from five consecutive acquisition cycles. Characteristic gas concentration data are processed using an adaptive filtering method to filter out concentration abrupt changes. The 3σ criterion is used to identify abnormal data. For occasional abnormal values, linear interpolation of two adjacent valid data is used for replacement. For abnormal values ​​that occur in three or more consecutive periods, they are determined to be sensor faults. Based on the added unified timestamp, data from the same time point collected by different sensors are matched, and linear interpolation is used to supplement the missing small amount of data. The min-max normalization method is used to map all preprocessed data to the [0,1] interval to eliminate different dimensions.

4. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The process of extracting the time-domain and phase characteristics of ultrasonic signals, relevant parameters of dielectric loss and leakage current, temperature and humidity differences and rates of change, and characteristic gas concentrations and ratios of insulating oil based on preprocessed multi-source data, and then removing redundant features to form a single-sensor insulation feature set, specifically includes: For ultrasonic partial discharge data, time-domain features and phase features are extracted. For dielectric loss data, real-time dielectric loss value, rate of change of dielectric loss value, and amplitude of dielectric loss value fluctuation are extracted. For leakage current data, peak current value, effective current value, harmonic components of current, and current change trend are extracted. For temperature and humidity data, ambient humidity value, equipment surface temperature value, temperature difference value, and rate of change of temperature and humidity are extracted. For characteristic gas data of insulating oil, concentration values ​​of methane, ethane, ethylene, and acetylene and the rate of change of each gas concentration are extracted, and gas concentration ratio parameters are also extracted. All extracted feature parameters were validated for effectiveness, and redundant features that were not related to the insulation state were removed to form a single-sensor insulation feature set.

5. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The process of combining weighted fusion and DS evidence theory to assign weights to the feature sets of each single sensor, obtaining preliminary fused features through weighted summation, eliminating information redundancy and conflicts through evidence synthesis, and obtaining a comprehensive insulation feature set after normalization specifically includes: Based on the measurement accuracy, data reliability, and correlation between characteristic parameters and insulation status of each sensor, the weights of the feature sets of each individual sensor are obtained. The corresponding feature parameters in each individual sensor feature set are weighted and summed according to the assigned weights to obtain preliminary fused features; The DS evidence theory is used to perform secondary fusion of the initial fusion features, treating each initial fusion feature as independent evidence, determining the basic probability allocation function of each piece of evidence, and synthesizing all the evidence through evidence synthesis rules; The fused feature set is normalized to obtain the final integrated insulation feature set.

6. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The establishment of a four-level insulation status classification standard, combined with a comprehensive feature set, operating standards, and historical data, employs a fuzzy comprehensive evaluation model. Quantitative evaluation is performed through membership functions and weight calculations to obtain the current insulation level of the equipment. Specifically, this includes: The insulation condition is divided into four levels: normal condition, warning condition, abnormal condition, and critical condition. A delay determination mechanism is set up so that when the comprehensive feature parameter is continuously lower or higher than the corresponding threshold for more than 60 seconds, it is determined to be the corresponding state. A fuzzy comprehensive evaluation model is used for hierarchical evaluation. The comprehensive insulation feature set is used as the evaluation factor set, and the four-level insulation status standard is used as the evaluation comment set. The membership function of each evaluation factor is calculated. The weights of each evaluation factor are determined by the analytic hierarchy process (AHP), and fuzzy synthesis is performed to obtain a comprehensive evaluation result of the equipment insulation status and output the current insulation status level of the equipment.

7. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The four-level early warning system is based on the four-level insulation status. Level 1 only records the status; Level 2 triggers a yellow alert and is included in the inspection; Level 3 triggers an orange audible and visual alarm and pushes information; Level 4 triggers a red alert, isolates the equipment, and sends an emergency alert. All early warning actions are logged in their entirety, including: Corresponding to the four-level insulation status standard, four levels of early warning are set. Level 1 early warning corresponds to the normal state and does not trigger any early warning action; Level 2 early warning corresponds to the attention state, triggers a yellow warning, records the early warning log, and incorporates the early warning information into the inspection; Level 3 early warning corresponds to the abnormal state, triggers an orange audible and visual alarm, points out the faulty equipment and abnormal characteristic parameters; Level 4 early warning corresponds to the critical state, triggers the highest level red audible and visual alarm, physically isolates the faulty equipment through the linkage equipment protection system, and pushes the fault handling plan. All warning actions are logged, including the warning trigger time, warning level, abnormal parameters, and executed actions.

8. The method for assessing and warning the insulation status of substation equipment using multi-sensor fusion according to claim 1, characterized in that, The process of collecting on-site verification and handling data, determining the accuracy of early warnings, optimizing evaluation model parameters and feature extraction for false and missed early warnings, and optimizing sensor deployment and early warning mechanisms through regular analysis of historical data specifically includes: After on-site verification, the verification results are entered into the data processing terminal. By sorting and analyzing the feedback data, the accuracy of the warning trigger and the rationality of the warning level are judged. For false alarms, analyze the causes of false alarms, adjust the evaluation threshold, membership function or feature weights, and optimize the evaluation model. For leakage warnings, supplement fault data, add relevant characteristic parameters, regularly conduct statistical analysis on historical warning data and fault data, generate insulation fault analysis reports, and optimize sensor deployment locations, acquisition frequencies, and warning mechanisms.

9. A multi-sensor fusion-based substation insulation condition assessment and early warning system, used to implement the multi-sensor fusion-based substation insulation condition assessment and early warning method as described in any one of claims 1-8, characterized in that, include: Data acquisition and preprocessing module: Distributed deployment of multiple sensors to perform layered processing on raw data, including noise reduction, outlier identification and processing, data alignment and missing value filling, and normalization; Feature extraction and filtering module: Extracts various insulation feature parameters from the preprocessed data, removes redundant features, and forms a single sensor feature set; Multi-source information fusion module: Combining weighted fusion and DS evidence theory, it performs weight allocation, preliminary fusion and evidence synthesis on single sensor features to generate a comprehensive insulation feature set; Condition assessment and classification module: Based on the fuzzy comprehensive evaluation model, combined with operating standards and historical data, calculates the membership degree and level of insulation condition; Early warning and logging module: Triggers corresponding early warnings based on insulation level, performs actions such as recording, inspection notification, audible and visual alarms or equipment isolation, and logs the entire process.

10. Model Optimization and Feedback Module: Collects on-site verification data, analyzes the accuracy of early warnings, and optimizes and evaluates model parameters, feature extraction strategies, and sensor deployment; Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

Images