Air switch on-line monitoring system and method integrated with multi-parameter sensor

Abstract

The invention discloses an air switch on-line monitoring system and method fused with a multi-parameter sensor, and belongs to the technical field of electrical equipment state monitoring, and the method comprises the steps: obtaining a multi-parameter real-time data flow of an air switch, carrying out the dynamic correction of an initial alarm threshold value through combining with humidity data, and generating an environment self-adaptive threshold value; current fluctuation features and vibration spectrum features are extracted, the temperature change rate is calculated, and a multi-dimensional fault feature set is generated; carrying out environment adaptability correction by combining an environment adaptive threshold value, and fusing to generate a composite fault index; and obtaining equipment aging data for aging effect compensation, generating a final state evaluation value, comparing the final state evaluation value with the judgment reference, and outputting a fault early warning signal when a trigger condition is met. According to the method, the technical means of fusing multi-dimensional features and combining environmental adaptability correction and aging effect compensation is adopted, real fault signals can be accurately stripped from the environment and aging interference, and therefore accurate early warning of potential faults of equipment is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical equipment state monitoring, in particular to an air switch online monitoring system and method fusing a multi-parameter sensor. BACKGROUND

[0002] An air switch, also known as a low-voltage circuit breaker, is a core protection electrical appliance widely used in power distribution networks, which undertakes the important functions of connecting, carrying and breaking normal circuit current and reliably cutting off the circuit under fault conditions such as overload and short circuit. The reliability of its operating state is directly related to the safety and stability of the entire power system, so it is of great significance to conduct real-time and effective online monitoring of the air switch.

[0003] In the prior art, the monitoring means for air switches are relatively limited. Some schemes rely on periodic offline detection or manual inspection, which is poor in timeliness and cannot capture transient abnormalities. Some online monitoring methods only monitor a single electrical parameter, such as monitoring the line current with a current transformer to achieve overcurrent protection, or only using infrared temperature measurement technology to monitor the temperature of the terminal, to determine whether there is poor contact. These methods can to some extent discover some more obvious faults.

[0004] However, the above prior art solutions have obvious technical defects. The monitoring method that relies solely on a single physical quantity is insufficient in information dimension and cannot fully reflect the complex operating state of the air switch. Early weak faults caused by the coupling of multiple factors, such as slight insulation degradation or internal arc precursors, are often not effectively identified. In addition, the existing monitoring systems mostly use fixed and static alarm thresholds, without considering the influence of changes in operating environment such as temperature and humidity on the insulation performance of the equipment and the fault triggering threshold, which can easily result in missed reports in harsh environments or false reports in normal environments. At the same time, the existing technology also generally ignores the natural drift of performance parameters due to normal aging and wear during long-term operation of the equipment, often misjudging this normal decline as a fault, affecting the accuracy of diagnosis. SUMMARY

[0005] To solve the above problems, the present application provides an air switch online monitoring system and method fusing a multi-parameter sensor, which uses a technical means that fuses multi-dimensional features and combines environmental adaptability correction and aging effect compensation, to accurately separate real fault signals from environmental and aging interference, thereby achieving accurate early warning of potential faults of the equipment.

[0006] The above object can be achieved by the following scheme: The air switch online monitoring system and method of fusing multi-parameter sensors comprise: obtaining current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of an air switch to obtain a multi-parameter real-time data stream; obtaining an initial alarm threshold for defining a normal range of current and temperature, and dynamically correcting the initial alarm threshold in combination with humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; extracting current fluctuation features and vibration spectrum features from the multi-parameter real-time data stream, and calculating a temperature change rate to generate a multi-dimensional fault feature set; combining the environment adaptive threshold with the multi-dimensional fault feature set for environment adaptability correction, and fusing the corrected multi-dimensional fault feature set to generate a composite fault index; obtaining device aging data in historical fault data to determine parameter drift rules, and compensating the composite fault index according to the rules to generate a final state evaluation value; obtaining a judgment benchmark for triggering a warning, and comparing the final state evaluation value with the judgment benchmark, and outputting a fault warning signal when a triggering condition is met.

[0007] Optionally, the obtaining of the multi-parameter real-time data stream comprises: synchronously high-frequency sampling the current sensor data, the temperature sensor data, the humidity sensor data and the vibration sensor data to obtain original multi-channel data; applying a digital filter to the original multi-channel data to filter out environmental noise to obtain pure signal data; and performing normalization processing on the pure signal data to eliminate dimension differences to obtain the multi-parameter real-time data stream.

[0008] Optionally, the generating of the environment adaptive threshold comprises: separating humidity data from the multi-parameter real-time data stream, and generating a humidity compensation factor according to a corresponding relationship between the humidity data and the compensation factor; and applying the humidity compensation factor to multiply the initial alarm threshold to reduce the threshold in a high-humidity environment to generate the environment adaptive threshold.

[0009] Optionally, the generating of the multi-dimensional fault feature set comprises: performing time-domain statistical analysis on the current sensor data in the multi-parameter real-time data stream to calculate a peak factor, a waveform margin and a standard deviation and combine them into a vector as current fluctuation features; performing fast Fourier transform on the vibration sensor data in the multi-parameter real-time data stream to obtain a vibration spectrum, identifying specific high-frequency band energy distribution associated with arc fault from the vibration spectrum and quantifying the specific high-frequency band energy distribution as vibration spectrum features; calculating a temperature change rate of the temperature sensor data in the multi-parameter real-time data stream as a temperature dynamic feature; and combining the current fluctuation features, the vibration spectrum features and the temperature dynamic feature to generate the multi-dimensional fault feature set.

[0010] Optionally, the generating the composite fault index comprises: performing environmental adaptability correction on the multi-dimensional fault feature set in combination with the environment-adaptive threshold to obtain a corrected multi-dimensional fault feature set; performing weighted fusion on current fluctuation features, vibration spectrum features and temperature dynamic features in the corrected multi-dimensional fault feature set to obtain a preliminary fault score; and performing non-linear correction on the preliminary fault score to generate the composite fault index.

[0011] Optionally, the generating the final state evaluation value comprises: establishing a quantitative correspondence relationship between a parameter drift amount and running time by analyzing device aging data in the historical fault data; calculating a corresponding normal aging drift amount according to the quantitative correspondence relationship and the current running time; and subtracting the normal aging drift amount from the composite fault index to generate the final state evaluation value.

[0012] Optionally, the obtaining the determination reference for triggering the early warning comprises: setting a basic early warning threshold according to a rated working parameter and an operating environment condition of the air switch; establishing a multi-level early warning grade division standard based on fault mode analysis in historical fault cases; dynamically adjusting the basic early warning threshold in combination with a current operating state of the device and the environment-adaptive threshold to obtain an adjusted early warning threshold; and mapping the adjusted early warning threshold to the multi-level early warning grade division standard to obtain the determination reference for triggering the early warning.

[0013] Optionally, the outputting the fault early warning signal comprises: performing real-time comparison between the final state evaluation value and the determination reference to determine whether a triggering condition of any early warning grade is reached, and if so, determining an early warning grade and outputting the fault early warning signal.

[0014] Optionally, the method further comprises: packaging the multi-parameter real-time data stream triggering the early warning signal and the composite fault index to form a fault event to be confirmed; receiving a manual verification result of the fault event to be confirmed to generate a labeled fault sample; and using the labeled fault sample to iteratively correct the quantitative correspondence relationship between the parameter drift amount and the running time to realize continuous optimization of the correspondence relationship.

[0015] Based on the same inventive concept, the application also provides an air switch online monitoring system with a fusion multi-parameter sensor, comprising: a data acquisition module for acquiring current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of the air switch to obtain a multi-parameter real-time data stream; an adaptive threshold module for acquiring an initial alarm threshold for defining a normal range of current and temperature, and dynamically correcting the initial alarm threshold in combination with humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; a feature extraction module for extracting current fluctuation features and vibration spectrum features from the multi-parameter real-time data stream, and calculating a temperature change rate to generate a multi-dimensional fault feature set; a feature fusion module for combining the environment adaptive threshold to perform environment adaptability correction on the multi-dimensional fault feature set, and fusing the corrected multi-dimensional fault feature set to generate a composite fault index; an aging compensation module for acquiring device aging data in historical fault data to determine a parameter drift rule, and compensating the composite fault index for aging effect according to the rule to generate a final state evaluation value; and a warning decision module for acquiring a decision benchmark for triggering a warning, and comparing the final state evaluation value with the decision benchmark, and outputting a fault warning signal when a triggering condition is met.

[0016] Compared with the prior art, the application has the following advantages: The application constructs a comprehensive perception system for the running state of the air switch by fusing multi-dimensional information such as current, temperature, humidity and vibration, overcomes the limitations of one-sided monitoring information of a single parameter and inability to effectively identify complex coupled faults, and significantly improves the accuracy and coverage of fault diagnosis.

[0017] The application innovatively introduces an environment adaptive threshold correction mechanism, can dynamically adjust the alarm benchmark according to real-time environmental humidity, so that the monitoring system can actively adapt to changes in the external environment, improve the monitoring sensitivity in high-risk working conditions such as high humidity, avoid false alarms in normal working conditions, and enhance the stability and environmental robustness of the entire monitoring system.

[0018] The application successfully separates the slow performance degradation caused by normal wear and tear and aging of the device from the sudden fault risk caused by abnormal working conditions by establishing a device aging model and compensating for the aging effect of the monitoring results, solves the problem of inaccurate diagnosis caused by parameter drift of long-running devices, and realizes accurate evaluation of the real health status of the device.

[0019] The closed-loop feedback learning mechanism designed in the application enables the monitoring system to have the ability of self-improvement and continuous learning through artificial verification of warning events and iterative optimization of model parameters, can continuously improve the cognitive accuracy of the device aging law and fault mode, and ensures the long-term effectiveness and high reliability of the monitoring method in the whole life cycle of the device.

[0020] Other features and advantages of the present application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The purposes and other advantages of the present application will be realized and attained by the structures described in the description, the claims, and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

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

[0022] Figure 1 is a flowchart of the air switch online monitoring method of the fusion multi-parameter sensor of the embodiment of the present application.

[0023] Figure 2 is a reference aging curve fitting schematic diagram of the embodiment of the present application.

[0024] Figure 3 is a final state evaluation value and early warning level threshold value comparison flowchart of the embodiment of the present application.

[0025] Figure 4 is a structural schematic diagram of the air switch online monitoring system of the fusion multi-parameter sensor of the embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purposes, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0027] With reference to Figure 1 , one embodiment of the present application proposes an air switch online monitoring method of fusion multi-parameter sensor, which adopts a technical means of fusing multi-dimensional features and combining environmental adaptability correction and aging effect compensation, can accurately strip the real fault signal from the environmental and aging interference, so as to realize accurate early warning of potential faults of the equipment.

[0028] The method of the embodiment specifically includes: Obtaining current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of the air switch to obtain a multi-parameter real-time data stream; Obtaining an initial alarm threshold for defining a normal range of current and temperature, and dynamically correcting the initial alarm threshold in combination with humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; Extracting current fluctuation features and vibration frequency spectrum features from the multi-parameter real-time data stream, and calculating a temperature change rate to generate a multi-dimensional fault feature set; Combining the environment adaptive threshold with the multi-dimensional fault feature set for environment adaptability correction, and fusing the corrected multi-dimensional fault feature set to generate a composite fault index; Obtaining device aging data in historical fault data to determine parameter drift rules, and compensating the composite fault index for aging effect according to the rules to generate a final state evaluation value; Obtaining a decision reference for triggering a warning, and comparing the final state evaluation value with the decision reference, and outputting a fault warning signal when the triggering condition is met.

[0029] Specifically, the method first obtains a comprehensive real-time data stream by fusing sensor data of four different physical dimensions of current, temperature, humidity and vibration. The core innovation is that it does not use a fixed threshold for judgment, but introduces a double dynamic correction mechanism. The first is environment adaptation, which dynamically corrects the basic alarm threshold using real-time humidity data to adapt to the influence of environmental changes on the insulation performance of the device. The second is aging effect compensation, which analyzes historical data to establish the parameter drift rules of normal aging of the device, and eliminates the predictable normal decay effect from the current comprehensive fault evaluation. The whole process from multi-dimensional data acquisition to feature extraction and fusion, to environment correction and aging compensation, finally generates a final state evaluation value that can accurately reflect the abnormal and sudden health risks of the device, and compares it with the decision reference to realize intelligent warning. The method can provide a highly reliable warning signal, provide a solid basis for predictive maintenance, effectively prevent fault escalation, and ensure the safe and stable operation of the power system.

[0030] Optionally, the obtaining a multi-parameter real-time data stream comprises: Synchronously high-frequency sampling the current sensor data, temperature sensor data, humidity sensor data and vibration sensor data to obtain original multi-channel data; Applying a digital filter to the original multi-channel data to filter out environmental noise to obtain pure signal data; Normalizing the pure signal data to eliminate dimensional differences to obtain a multi-parameter real-time data stream.

[0031] Specifically, first, through an integrated data acquisition device, the current sensor, temperature sensor, humidity sensor and vibration sensor installed at the key position of the air switch are synchronously high-frequency sampled. Synchronous sampling ensures that the current, temperature, humidity and vibration data collected at any time point are accurately corresponding, providing a time reference for subsequent analysis of the instantaneous correlation between different physical quantities. High-frequency sampling ensures that fault signs such as current mutation or high-frequency vibration caused by arc discharge can be captured. The original, unprocessed multi-channel time series data obtained in this process is the original multi-channel data. Subsequently, digital filtering techniques are applied to the original multi-channel data to filter out environmental noise interference. According to the signal characteristics of different sensors, appropriate filters are used. For example, for temperature and humidity signals that change relatively slowly, low-pass filters are used to filter out high-frequency random fluctuations; for current and vibration signals, band-pass filters are designed to retain specific frequency band signals related to fault characteristics, while filtering out power frequency interference and irrelevant mechanical vibration background noise. After filtering, the signal-to-noise ratio of the signal is significantly improved, obtaining a clear and pure signal data that can reflect the true state of the device. Finally, the pure signal data is normalized to eliminate differences caused by different physical quantities and numerical ranges, establishing a unified data basis for subsequent multi-dimensional feature fusion. This step can use the maximum and minimum normalization method, and the processing process is as follows: , wherein, represents the normalized data point. is a specific measurement value in the pure signal data. and are the maximum and minimum values of the corresponding sensor under normal operating conditions, preset or statistically historical. For example, for current data, is the upper limit of the rated current, is zero or a very small reference value. This normalization operation is applied independently to the current, temperature, humidity and vibration data channels, mapping their respective values to a unified, dimensionless interval, usually between 0 and 1. Thus, after integrating all standardized signal channels, a multi-parameter real-time data stream is generated for subsequent algorithms to use directly. This method realizes high-quality acquisition of multi-dimensional physical information of air switches through synchronous high-frequency sampling, digital filtering and normalization. This process not only ensures the accurate alignment of data in time, but also greatly improves the quality and usability of raw data by removing noise and eliminating dimensional effects.

[0032] Optionally, the generating an environment adaptive threshold comprises: separating humidity data from the multi-parameter real-time data stream, and generating a humidity compensation factor according to the corresponding relationship between the humidity data and the compensation factor; The initial alarm threshold is multiplied by the humidity compensation factor to reduce the threshold in a high humidity environment, generating an environment adaptive threshold.

[0033] Specifically, first, the real-time humidity data is separated from the obtained multi-parameter real-time data stream. At the same time, an initial alarm threshold for defining the normal working range of current and temperature is set in advance, which is a benchmark established based on the normal operating characteristics of the device in a standard, dry environment. Then, a humidity compensation factor is generated according to the preset corresponding relationship between the humidity data and the compensation coefficient.

[0034] The corresponding relationship is obtained by establishing a lookup table. The air switch is placed in an environment test box with accurately controllable temperature and humidity, and a standard dry environment is set as the benchmark condition to measure the key electrical insulation threshold, denoted as , and the compensation coefficient under this benchmark humidity is defined as 1.0. Subsequently, the temperature and other conditions are kept unchanged, and the environmental humidity is gradually increased to multiple preset gradient points. At each humidity gradient point, the above-mentioned electrical insulation threshold is repeatedly measured to obtain . The compensation coefficient at this humidity point is calculated as . Since the increase of humidity will lead to the decrease of insulation performance ( ), the compensation coefficient is a non-dimensional value less than or equal to 1.0. All humidity gradient points and their corresponding compensation coefficients are stored in a lookup table in the form of "humidity-compensation coefficient" key-value pairs. After obtaining real-time humidity data , first, a query is made in the lookup table. If the real-time humidity value matches a certain humidity gradient in the lookup table, the corresponding compensation coefficient is directly used as the current humidity compensation factor. If is between two adjacent humidity gradients and in the lookup table (their corresponding compensation coefficients are and respectively), a linear interpolation method is used to calculate the accurate humidity compensation factor . The linear interpolation calculation formula is . After obtaining the humidity compensation factor, the initial alarm threshold is dynamically corrected using the factor. This correction process is realized by multiplication operation, and the mathematical expression is as follows: , wherein is the final generated environment adaptive threshold; represents the preset initial alarm threshold. Since the value of is less than 1 in a high humidity environment, the multiplication operation will result in below , so as to effectively reduce the alarm threshold in high humidity environment. Conversely, in dry environment, close to or equal to 1, the threshold will remain at or close to the initial level. This process is carried out in real time, so that the alarm threshold can be continuously dynamically adjusted with the change of environmental humidity. The method introduces a humidity compensation mechanism, so that the alarm threshold of the monitoring system is no longer fixed, but can actively adapt to the change of environmental humidity, significantly improving the accuracy and sensitivity of monitoring.

[0035] Optionally, the generating the multi-dimensional fault feature set comprises: performing time domain statistical analysis on the current sensor data in the multi-parameter real-time data stream, calculating its peak factor, waveform margin and standard deviation and combining them into a vector as the current fluctuation feature; performing fast Fourier transform on the vibration sensor data in the multi-parameter real-time data stream to obtain a vibration frequency spectrum, identifying and quantifying the specific high-frequency band energy distribution associated with arc fault from the vibration frequency spectrum as the vibration spectrum feature; calculating the temperature change rate of the temperature sensor data in the multi-parameter real-time data stream as the temperature dynamic feature; combining the current fluctuation feature, vibration spectrum feature and temperature dynamic feature to generate the multi-dimensional fault feature set.

[0036] Specifically, first, time-domain statistical analysis is performed on the current sensor data in the multi-parameter real-time data stream. This analysis is performed within a pre-set time window, and several statistical indicators are calculated to quantify the changes in the current waveform. Specifically, the peak factor, which is the ratio of the peak value of the current signal to the effective value, is calculated to measure the impact of the waveform; the waveform margin, which reflects the deviation of the waveform from the standard sine wave, is calculated; and the standard deviation, which represents the dispersion and instability of the current signal within a cycle, is calculated. These three indicators together constitute a quantitative description of the current abnormal disturbance and are combined into a vector as the current fluctuation feature. Second, frequency spectrum analysis is performed on the vibration sensor data in the multi-parameter real-time data stream. By applying the Fast Fourier Transform, the time-domain vibration signal is converted to the frequency-domain vibration spectrum. When an arc fault occurs inside the air switch, a specific high-frequency vibration will be generated due to the instantaneous and violent expansion of the gas. Therefore, in the obtained vibration spectrum, the specific high-frequency band closely related to the arc fault determined through experiments or prior knowledge is evaluated for energy. By conducting simulated fault experiments on the same type or same type of air switch in a laboratory environment, artificially introducing arc faults of different intensities, and synchronously collecting their vibration signals, the vibration spectrum under normal operation and fault conditions is compared to identify the frequency band with significantly enhanced and stable energy, thereby determining the characteristic frequency band under this fault mode. Then, the total signal energy or energy proportion in this frequency band is calculated, quantifying the complex spectrum information into a single numerical value, which is the vibration spectrum feature that can effectively indicate potential arc discharge activity. Third, dynamic characteristic analysis is performed on the temperature sensor data in the multi-parameter real-time data stream. Unlike focusing on the absolute temperature, this method focuses on calculating the temperature change rate to capture the dynamic evolution of the device's thermal state. The formula for calculating the temperature change rate is as follows: , wherein, represents the temperature change rate, is the temperature measurement value at the current time, is the temperature value at the previous sampling time, is the time interval between the two sampling points. This temperature change rate, as an independent feature, the temperature dynamic feature, is particularly sensitive to rapid temperature rise phenomena caused by poor contact, overload, or arc. Finally, the current fluctuation feature vector, vibration spectrum feature scalar, and temperature dynamic feature scalar obtained in the above steps are combined to form a comprehensive multi-dimensional fault feature set, providing a comprehensive state representation for subsequent fault diagnosis. This multi-dimensional and multi-angle feature extraction strategy greatly enhances the sensitivity and accuracy of fault recognition, effectively distinguishes different fault modes, and provides a solid technical foundation for the early detection of weak faults, improving the predictability and reliability of the entire monitoring system.

[0037] Optionally, the generating the composite fault index comprises: environmentally correcting the multi-dimensional fault feature set in combination with the environment adaptive threshold to obtain a corrected multi-dimensional fault feature set; weighting and fusing the current fluctuation feature, the vibration spectrum feature and the temperature dynamic feature in the corrected multi-dimensional fault feature set to obtain a preliminary fault score; nonlinearly correcting the preliminary fault score to generate the composite fault index.

[0038] Specifically, the current fluctuation feature, the vibration spectrum feature and the temperature dynamic feature extracted from the multi-parameter real-time data stream are compared and normalized by the environment adaptive threshold respectively. This correction process converts the original feature values into the relative abnormality degree relative to the current environmental conditions, thereby obtaining the corrected multi-dimensional fault feature set. This step ensures that the subsequent evaluation is not directly disturbed by environmental factors, so that the feature values can more purely reflect the health status of the device itself. Subsequently, each feature component in the corrected multi-dimensional fault feature set is weighted and fused to calculate a preliminary fault score. The fusion process aims to integrate information from different physical dimensions into a single index, and the calculation expression is as follows: , wherein, represents the preliminary fault score; , and are the corrected current fluctuation feature, the vibration spectrum feature and the temperature dynamic feature, which are dimensionless relative values; , and are preset weight coefficients, which are also dimensionless, and the sum is usually 1. These weight coefficients are determined based on expert experience or historical data analysis, reflecting the contribution difference of different features to the indication of the overall fault risk of the air switch. For example, the vibration spectrum feature directly related to the electric arc may be given a higher weight. Finally, the calculated preliminary fault score is nonlinearly corrected to generate the final composite fault index. This step aims to amplify the synergistic effect of the fault, that is, when multiple features are abnormal at the same time, the comprehensive risk is greater than the simple linear superposition of each part. By applying a preset nonlinear mapping function, for example, the preliminary fault score is transformed by a function such as to generate the composite fault index : , wherein, is the center point offset of the function, representing a critical transition point of the preliminary failure score, whose value is set according to historical data or expert experience, for example, it can be set to 0.5, indicating When it exceeds 0.5, the risk of failure begins to accelerate accumulation. is the gain coefficient, used to control the steepness of the curve, that is, the sensitivity of risk amplification. A larger value will cause a sharp jump near , suitable for scenarios that require high sensitivity alarm; a smaller value provides a smoother transition. By setting these two parameters, the non-linear amplification rule of risk can be accurately defined, ensuring the reproducibility of the composite failure index generation process. After this series of processing, the composite failure index is finally obtained, which can comprehensively and dynamically represent the current failure risk level of the air switch. This method realizes an intelligent conversion from multi-dimensional features to a single risk index through environmental adaptability correction, weighted fusion and non-linear correction. It not only integrates the information of multiple sensors to form a comprehensive state evaluation, but also excludes the interference of external environment through environmental self-adaptive correction, making the evaluation result more objective and accurate.

[0039] Optionally, the generating the final state evaluation value comprises: establishing a quantitative correspondence between the parameter drift and the running time by analyzing the equipment aging data in the historical failure data; calculating the corresponding normal aging drift according to the quantitative correspondence and the current running time; subtracting the normal aging drift from the composite failure index to generate the final state evaluation value.

[0040] Specifically, this process utilizes a large amount of full life cycle monitoring data of the same type of air switch from the time of putting into use to the end of life. By performing polynomial regression analysis modeling on the trajectory of the natural growth of the composite failure index with the running time in these data, a baseline aging curve is extracted. This curve accurately describes the quantitative correspondence between the predictable and slow drift of the composite failure index due to normal aging factors such as contact wear and insulation material degradation and the running time of the equipment under the condition of no any sudden failure. The specific method of establishing the quantitative correspondence is: first, the monitoring data of the equipment running under the condition of no sudden failure is selected from the historical database to form a data point set composed of running time and the corresponding composite failure index . Secondly, a polynomial regression model is used to fit the data set to establish the functional relationship between the aging drift and the running time , and a second-order polynomial model is selected: Then, the least square method is applied as the optimization algorithm to calculate the optimal coefficients , which minimize the mean square error between the model prediction and the actual composite failure index in the dataset. Finally, the goodness of fit of the model is evaluated by calculating the coefficient of determination , which ensures that the model can accurately describe the aging trend. As shown in Figure 2 , based on the historical monitoring sample points of the same type of air switch under the condition of no sudden failure, a second-order polynomial regression is used to fit the natural growth trajectory of the composite failure index with the running time, and the baseline aging curve is obtained. Next, according to the established quantitative correspondence and the current cumulative running time of the air switch, the normal aging drift amount corresponding to the current time is calculated. The current running time can be obtained from the device management system, with the unit of cumulative working hours or switch action times. By substituting this running time into the aforementioned aging drift model, a specific numerical value, i.e., the normal aging drift amount, can be calculated. This drift amount has the same dimension and scale as the composite failure index, representing the "background" risk value contributed by normal decay at the current device age. Finally, the normal aging drift amount calculated is subtracted from the composite failure index, thereby generating the final state evaluation value. This calculation process can be represented by the following formula: , In this formula, is the final state evaluation value, represents the composite failure index containing all information of aging and potential failure, is the normal aging drift amount calculated according to the current running time. The essence of this operation is to remove the part of the composite failure index caused by normal and predictable aging effects, thereby highlighting the real failure risk caused by unexpected and abnormal conditions. This method realizes dynamic baseline correction of device state evaluation by introducing an aging effect compensation mechanism, which can effectively distinguish between slow performance degradation caused by normal wear and tear and real failure signs caused by sudden or accelerated abnormalities.

[0041] Optionally, the obtaining of the determination benchmark for triggering the early warning comprises: setting a basic early warning threshold according to the rated working parameters and operating environment conditions of the air switch; establishing a multi-level early warning grade division standard based on failure mode analysis in historical failure cases; combining the current operating state of the device and the environment-adaptive threshold to dynamically adjust the basic early warning threshold to obtain an adjusted early warning threshold; mapping the adjusted early warning threshold with the multi-level early warning grade division standard to obtain the determination benchmark for triggering the early warning.

[0042] Specifically, to obtain the criteria for triggering early warnings, a static basic early warning threshold is first set based on the technical specifications of the air switch and its preset operating environment conditions. This threshold represents the safety boundary that the various state parameters of the equipment should not exceed under ideal operating conditions. Its value comes from the rated operating parameters of the equipment, such as rated current and maximum allowable temperature rise, and is a conservative and fixed reference value. Simultaneously, based on in-depth analysis of historical fault cases, a multi-level early warning classification standard is established. By studying a large amount of past fault data, the characteristic signal combinations and their severity corresponding to different fault modes are identified. Accordingly, the early warning response is divided into multiple levels, such as "attention level," "early warning level," and "severe level." Each level corresponds to a numerical range consisting of the final state assessment value, thus forming a structured fault severity assessment framework. Next, the aforementioned basic early warning threshold is dynamically adjusted to generate an adjusted early warning threshold. The specific algorithm for dynamic adjustment is as follows: First, a load rate compensation factor is calculated. Obtain real-time current values ​​from multi-parameter real-time data streams. And in conjunction with the rated current of the equipment Calculate the current load rate Then, calculations are performed according to the preset step linearity rule. :when When the load is below the set high load threshold, The value is 1.0, and no further adjustments are made; when When the load is greater than or equal to the high load threshold, Decreasing linearly, its calculation formula is as follows: ,in Let be the load sensitivity coefficient, and The minimum value is set to prevent over-adjustment. Then, the environmental adaptive threshold generated in the previous steps is... Multiplying this by the load rate compensation factor yields the final adjusted warning threshold. Finally, the dynamically generated and adjusted warning thresholds are mapped to a pre-established multi-level warning classification standard to obtain the final judgment criterion. This mapping process uses the adjusted warning thresholds as the upper limit of 100% risk, and then calculates the specific trigger values ​​for each warning level according to the percentages or relative values ​​defined in the multi-level warning classification standard. The final output judgment criterion is not a single value, but a structured set containing multiple trigger conditions of different severity levels, providing a clear and hierarchical basis for subsequent warning decisions. This method significantly improves the intelligence and practicality of the warning system by constructing a dynamic and multi-level judgment criterion. It abandons the traditional "one-size-fits-all" approach of fixed thresholds, enabling the warning strategy to proactively adapt to changing operating environments and equipment conditions, thereby ensuring sensitivity under high-risk conditions while effectively avoiding false alarms under normal operating conditions.

[0043] Optionally, the output fault warning signal includes: The final state evaluation value is compared with the judgment benchmark in real time to determine whether it meets the triggering conditions of any warning level. If it does, the warning level is determined and a fault warning signal is output.

[0044] Specifically, the final state assessment value obtained after aging effect compensation is first compared in real time with multiple warning level thresholds included in the judgment criterion. For example... Figure 3 As shown, the comparison process employs a top-down priority judgment logic: First, it checks whether the final state assessment value is greater than or equal to the highest-level "Severe" threshold; if so, the warning level is directly determined as "Severe." If not, it continues to check whether it is greater than or equal to the next lower-level "Warning" threshold; if so, the warning level is determined as "Warning." This process continues until all levels are judged. Once the final state assessment value meets the triggering condition for any warning level, the warning is immediately determined to be triggered, and a corresponding warning level identifier is generated. Subsequently, a structured fault warning signal is immediately generated and output. This signal is not just a simple alarm command, but a complete data packet containing information such as the current timestamp, device identifier, determined warning level identifier, and the final state assessment value that triggered the warning, facilitating subsequent analysis, recording, and response. This hierarchical judgment mechanism ensures that the device status can be accurately mapped to the highest relevant risk level and trigger an immediate response, greatly improving the predictability and efficiency of equipment maintenance.

[0045] Optionally, the method further includes: The multi-parameter real-time data stream that triggers the warning signal is packaged with the composite fault index to form a fault event to be confirmed. receiving a manual verification result of the to-be-verified fault event, and generating a labeled fault sample; Using the labeled fault sample, iteratively correcting the quantitative correspondence relationship between the parameter drift and the running time, and realizing continuous optimization of the correspondence relationship.

[0046] Specifically, when the fault early warning signal is output, a data packaging program will be automatically executed. The program will package the multi-parameter real-time data stream in the time period triggering the early warning and the composite fault index calculated at the early warning triggering moment into an independent digital archive. This archive is marked as a to-be-verified fault event and stored in the database, waiting for subsequent manual intervention. Next, the to-be-verified fault event will be submitted to the work interface of the operation and maintenance personnel or technical experts for manual verification. The operation and maintenance personnel will make a judgment on the nature of the event in combination with on-site investigation, device historical operation records and their own professional knowledge. The verification result can be to confirm it as a real fault, and further label the specific type of the fault, such as poor contact or arc fault; or determine it as a false alarm, which may be caused by external strong interference or temporary inaccuracy of the model. The system will receive this explicit manual verification result and associate it with the original event data package to form a labeled fault sample containing original data and expert conclusion. Finally, the method uses these continuously accumulated labeled fault samples to iteratively correct the quantitative correspondence relationship between the parameter drift and the running time. This correction process is an online learning or periodic retraining process. When a sample is labeled as a real fault, a significant increase in the composite fault index at that time is mainly due to abnormal events, rather than normal aging. Conversely, if it is labeled as a false alarm and other disturbances are excluded, it may mean that the normal aging drift at the current running time is underestimated. By using these labeled fault samples as new training data, updating the aging drift model, recalibrating the parameters, and enabling it to more accurately fit the performance degradation law of the device in the real world, the method realizes continuous optimization of the quantitative correspondence relationship. The method introduces a closed-loop manual verification and model self-optimization mechanism to build an intelligent monitoring system with learning and evolution capabilities. The system can obtain valuable feedback information from each early warning event, whether it is a real fault or a false alarm, to continuously refine its understanding of the normal aging process of the device.

[0047] Based on the same inventive concept, as Figure 4 The application also provides an air switch online monitoring system fused with a multi-parameter sensor, which comprises: A data acquisition module is configured to acquire current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of the air switch to obtain a multi-parameter real-time data stream. an adaptive threshold module configured to obtain an initial alarm threshold for defining a normal range of current and temperature, and dynamically correct the initial alarm threshold in combination with humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; a feature extraction module configured to extract current fluctuation features and vibration spectrum features from the multi-parameter real-time data stream, and calculate a temperature change rate to generate a multi-dimensional fault feature set; a feature fusion module configured to perform environment adaptability correction on the multi-dimensional fault feature set in combination with the environment adaptive threshold, and fuse the corrected multi-dimensional fault feature set to generate a composite fault index; an aging compensation module configured to obtain device aging data in historical fault data to determine a parameter drift rule, and compensate the composite fault index for aging effects according to the rule to generate a final state evaluation value; a warning decision module configured to obtain a decision benchmark for triggering a warning, and compare the final state evaluation value with the decision benchmark, and output a fault warning signal when a triggering condition is met.

[0048] To verify the feasibility of the present application in implementation, the present application is applied to an online monitoring system of a certain city power grid substation. The substation undertakes the power supply task of the key area, and the stable operation of the high-voltage air switch is crucial. The traditional monitoring method relies on regular inspection and fixed threshold alarm, which is difficult to cope with environmental changes such as humid weather, and cannot effectively distinguish between normal aging and real fault signs of the device, and there is a risk of false negatives and false positives. The present application aims to improve the accuracy and predictability of monitoring by fusing multi-parameter sensor data and introducing a dynamic correction mechanism.

[0049] In this embodiment, the online monitoring system of the present application is deployed on a high-voltage air switch that has been in operation for many years. Through the data acquisition module, the current sensor, temperature sensor, humidity sensor and vibration sensor installed at the key positions of the switch are synchronously sampled at high frequency, and after digital filtering and normalization processing, a multi-parameter real-time data stream is generated.

[0050] To verify the effectiveness of the present application, the air switch is continuously monitored for several months, the operation data under different weather conditions and load conditions are recorded, and the adaptive threshold generation, aging effect compensation and graded warning functions of the system are tested.

[0051] In the Meiyu season of June 2024, the substation environment humidity continues to be above 90%. At this time, the adaptive threshold module separates the humidity data from the multi-parameter real-time data stream and generates a humidity compensation factor of 0.85 according to the preset correspondence between humidity and compensation factor. Apply the factor to multiply the initial alarm threshold for defining current and temperature to generate an environment adaptive threshold, which is 15% lower than the initial value. This adjustment makes the system more sensitive in a humid high-risk environment and can capture weak anomalies caused by a decline in insulation performance.

[0052] In a load switching operation, the feature extraction module successfully extracts a multi-dimensional fault feature set from the multi-parameter real-time data stream. Specifically, the current sensor data shows that its peak factor and standard deviation have a transient increase, forming a current fluctuation feature. At the same time, the vibration sensor data, after fast Fourier transform, identifies significant energy aggregation in a specific high-frequency band associated with arc faults, quantified as a vibration spectrum feature. The temperature sensor records that the contact temperature has a temperature change rate that exceeds the normal range.

[0053] Subsequently, the feature fusion module first combines the environment adaptive threshold to correct the above multi-dimensional fault feature set for environmental adaptability, and then weights and fuses the corrected current fluctuation feature, vibration spectrum feature, and temperature dynamic feature for nonlinear correction to generate a composite fault index. For this air switch that has been running for 8 years, the aging compensation module calculates the normal aging drift amount corresponding to the current running time according to the parameter drift law established by the second-order polynomial regression analysis modeling based on its historical data, and subtracts it from the composite fault index to finally generate a final state evaluation value that accurately reflects the true health status of the device.

[0054] During the monitoring period, an event with a final state evaluation value of 0.36 was recorded. The early warning decision module first obtains the dynamic decision benchmark. At the time of this event, based on high humidity and high load, the adjusted early warning threshold of 0.40 is calculated as a 100% risk level. According to the preset multi-level division standard (for example, the "attention level" is The trigger value of the "attention level" is 0.32, the trigger value of the "early warning level" is 0.40, and the trigger value of the "serious level" is 0.60. Then, by using the top-down priority judgment logic, it is found that the final state evaluation value 0.36 does not reach the threshold values of the "serious level" and the "early warning level", but exceeds the trigger value 0.32 of the "attention level". Therefore, it is determined that the early warning is triggered, and the early warning level is identified as the "attention level". After the early warning is triggered, the multi-parameter real-time data stream at the triggering time and the composite fault index are packaged as a fault event to be confirmed. After the on-site verification of the operation and maintenance personnel, it is confirmed that the event is a slight loosening of the internal connecting part of the switch due to long-term vibration, rather than a serious fault. The artificial verification result is input into the system as a labeled fault sample, which is used to iteratively correct the quantitative correspondence relationship describing the parameter drift law, and realizes the closed-loop optimization of the model.

[0055] Through the present application, the state monitoring capability of the air switch of the substation is significantly improved. The system not only can adapt to environmental changes to dynamically adjust the monitoring sensitivity, but also can accurately distinguish between normal aging and real faults, realizes the transformation from "after-maintenance" to "predictive maintenance", and effectively prevents potential power accidents.

[0056] Table 1 Comparison table of air switch monitoring data and state evaluation

[0057] Table 2 Air switch fault early warning event record and response table

[0058] The above table 1 and table 2 record the actual application data of the present application in the monitoring of the air switch of the substation.

[0059] As can be seen from table 1, the aging effect compensation mechanism of the present application has significant effect. For example, on June 28, 2024, under the combined influence of high humidity and normal aging, the composite fault index reaches 0.62, and if the traditional method is used, a false alarm may be triggered due to exceeding the fixed threshold. The present application subtracts the normal aging drift amount of 0.26 to obtain the final state evaluation value of 0.36, which more accurately reflects the real risk caused by the slight abnormality of the device, and avoids unnecessary shutdown maintenance.

[0060] Table 2 demonstrates the grading early warning and closed-loop learning capability of the present application. For different severity faults, the present application can output differentiated early warning levels. For example, for slight looseness with a final state evaluation value of 0.36, a "attention level" early warning is issued, providing sufficient response and troubleshooting time for the operation and maintenance personnel. For severe ablation with an evaluation value of 0.82, a "serious level" early warning is triggered directly, guiding the operation and maintenance personnel to take decisive measures. At the same time, the result of each manual verification becomes new data for the optimization model, making the system's judgment more and more accurate, proving the great advantages of the present application in improving monitoring accuracy and achieving predictive maintenance.

[0061] It should be noted that the electrical connection between the above-mentioned units does not necessarily represent the direct connection of the line, and the indirect connection mode can be applied to the embodiments of the present application as long as the purpose of the present application is achieved. The above-described is only an exemplary embodiment of the present application, and cannot limit the scope of the present application.

[0062] That is, any equivalent changes and modifications made in accordance with the teachings of the present application are still within the scope of the present application. Other embodiments of the present application will be readily apparent to those skilled in the art upon considering the description and the disclosure of the true principles of the present application. The present application is intended to cover any variations, uses or adaptive changes of the present application that follow the general principles of the present application and include common knowledge or conventional technical means in the art that are not described in the present application.

Claims

1. A method for online monitoring of air break switch with fusion multi-parameter sensor, characterized in that, The method comprises: obtaining current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of the air switch to obtain a multi-parameter real-time data stream; obtaining an initial alarm threshold for defining a normal range of current and temperature, and dynamically correcting the initial alarm threshold in combination with humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; extracting current fluctuation features and vibration spectrum features from the multi-parameter real-time data stream, and calculating a temperature change rate to generate a multi-dimensional fault feature set; combining the environment adaptive threshold with the multi-dimensional fault feature set for environment adaptability correction, and fusing the corrected multi-dimensional fault feature set to generate a composite fault index; obtaining device aging data in historical fault data to determine parameter drift rules, and compensating the composite fault index according to the rules to generate a final state evaluation value; obtaining a decision reference for triggering a warning, and comparing the final state evaluation value with the decision reference, and outputting a fault warning signal when the triggering condition is met.

2. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 1, characterized in that, The multi-parameter real-time data stream comprises: synchronously high-frequency sampling the current sensor data, temperature sensor data, humidity sensor data and vibration sensor data to obtain original multi-channel data; applying a digital filter to the original multi-channel data to filter out environmental noise to obtain pure signal data; normalizing the pure signal data to eliminate dimension differences to obtain a multi-parameter real-time data stream.

3. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 2, characterized in that, The environment adaptive threshold comprises: separating humidity data from the multi-parameter real-time data stream, and generating a humidity compensation factor according to the corresponding relationship between humidity data and compensation factors; applying the humidity compensation factor to multiply the initial alarm threshold to reduce the threshold in a high humidity environment to generate an environment adaptive threshold.

4. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 2, characterized in that, The multi-dimensional fault feature set comprises: performing time domain statistical analysis on the current sensor data in the multi-parameter real-time data stream to calculate its peak factor, waveform margin and standard deviation and combine them into a vector as current fluctuation features; performing fast Fourier transform on the vibration sensor data in the multi-parameter real-time data stream to obtain a vibration spectrum, identifying and quantifying specific high-frequency band energy distribution associated with arc fault from the vibration spectrum as vibration spectrum features; calculating the temperature change rate of the temperature sensor data in the multi-parameter real-time data stream as a temperature dynamic feature; combining the current fluctuation features, vibration spectrum features and temperature dynamic features to generate a multi-dimensional fault feature set.

5. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 4, characterized in that, The composite fault index comprises: combining the environment adaptive threshold with the multi-dimensional fault feature set for environment adaptability correction to obtain a corrected multi-dimensional fault feature set; weighting and fusing the current fluctuation features, vibration spectrum features and temperature dynamic features in the corrected multi-dimensional fault feature set to obtain a preliminary fault score; nonlinearly correcting the preliminary fault score to generate a composite fault index.

6. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 5, characterized in that, The final state evaluation value comprises: establishing a quantitative corresponding relationship between parameter drift and running time by analyzing the device aging data in the historical fault data; According to the quantization correspondence and the current running time, a corresponding normal aging drift is calculated; The normal aging drift is subtracted from the composite failure index to generate a final state evaluation value.

7. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 1, characterized in that, The determination criterion for triggering the early warning includes: According to the rated operating parameters of the air switch and the operating environment conditions, a basic early warning threshold is set; Based on the failure mode analysis in the historical failure cases, a multi-level early warning grade division standard is established; The basic early warning threshold is dynamically adjusted in combination with the current operating state of the device and the environment adaptive threshold to obtain an adjusted early warning threshold; The adjusted early warning threshold is mapped with the multi-level early warning grade division standard to obtain the determination criterion for triggering the early warning.

8. The method of online monitoring of air break switch with fusion multi-parameter sensor according to claim 7, characterized in that, The output of the failure early warning signal includes: The final state evaluation value is compared with the determination criterion in real time to determine whether it reaches the triggering condition of any early warning grade, and if so, the early warning grade is determined and the failure early warning signal is output.

9. The method of online monitoring of air break switch according to claim 8, characterized in that, The method further includes: The multi-parameter real-time data stream triggering the early warning signal and the composite failure index are packaged to form a failure event to be confirmed; An artificial verification result of the failure event to be confirmed is received to generate a labeled failure sample; The quantization correspondence of the parameter drift with the running time is iteratively corrected using the labeled failure sample to realize continuous optimization of the correspondence.

10. The air switch online monitoring system of fusion multi-parameter sensor, applied to the air switch online monitoring method of fusion multi-parameter sensor as claimed in any one of claims 1-9, characterized in that, The system includes: A data acquisition module for acquiring current sensor data, temperature sensor data, humidity sensor data and vibration sensor data of the air switch to obtain a multi-parameter real-time data stream; An adaptive threshold module for obtaining an initial alarm threshold for defining the normal range of current and temperature, and dynamically correcting the initial alarm threshold in combination with the humidity data in the multi-parameter real-time data stream to generate an environment adaptive threshold; A feature extraction module for extracting current fluctuation features and vibration frequency spectrum features from the multi-parameter real-time data stream, and calculating a temperature change rate to generate a multi-dimensional failure feature set; A feature fusion module for combining the environment adaptive threshold to perform environment adaptability correction on the multi-dimensional failure feature set, and fusing the corrected multi-dimensional failure feature set to generate a composite failure index; An aging compensation module for acquiring device aging data in historical failure data to determine the parameter drift rule, and compensating the composite failure index for aging effect according to the rule to generate a final state evaluation value; An early warning decision module for obtaining a determination criterion for triggering the early warning, and comparing the final state evaluation value with the determination criterion, and outputting a failure early warning signal when the triggering condition is met.

Images