Fault prediction method and system for high-voltage switch operating mechanism

Abstract

The invention relates to the technical field of fault prediction, and particularly discloses a fault prediction method and system for a high-voltage switch operating mechanism. According to the method, early-stage accurate prediction of the fault of the high-voltage switch operating mechanism is realized through multi-feature extraction and fusion in combination with a lightweight Transformer model, and compared with a traditional single current signal analysis method, the method fully considers the coupling effect of electromagnetic and mechanical states, captures electromagnetic abnormality through current-electromagnetic coordination node features, and improves the accuracy of the fault diagnosis of the high-voltage switch operating mechanism. According to the method, mechanical vibration faults are recognized through vibration energy features, motion discordance is detected through electromagnetic driving force-motion acceleration cooperative features, so that possible fault modes of the mechanism are comprehensively covered, the accuracy and the real-time performance of evaluation are further improved through efficient fusion of the lightweight Transform model, the early fault missing report rate is effectively reduced, and the reliability of the mechanism is improved. And reliable guarantee is provided for safe operation of a power system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction, and in particular to a high-voltage switch operating mechanism fault prediction method and system. BACKGROUND

[0002] As the core control equipment for realizing circuit breaking and closing in the power system, the high-voltage switch operating mechanism is widely used in power plants, substations and power distribution networks and other key links. In the actual operating environment, the mechanism is usually installed in an outdoor substation or a sealed switch cabinet and is subjected to multiple stresses such as electromagnetic force impact, mechanical vibration, temperature change and environmental corrosion for a long time. Especially under harsh working conditions such as frequent load operation and short-circuit current impact, the core components such as springs, connecting rods and electromagnets inside the mechanism are prone to progressive failures such as fatigue damage, lubrication failure and jamming, which directly affects the reliability and safety of the power grid. However, the electromagnetic characteristics and mechanical state of the high-voltage switch operating mechanism are closely coupled in actual operation. When the transmission chain of the mechanism is slightly jammed, the current waveform changes weakly and is difficult to detect, but the mechanical vibration and movement displacement have obvious abnormalities. The existing fault prediction usually uses single current signal analysis, which can lead to a high early fault miss rate. Therefore, a high-voltage switch operating mechanism fault prediction method is needed to solve the above problems. SUMMARY

[0003] The purpose of the present application is to provide a high-voltage switch operating mechanism fault prediction method, comprising: obtaining operating data of a high-voltage switch operating mechanism within a preset time period, the operating data including current-derived features and mechanical feedback features; obtaining current waveform timing features and electromagnetic time-domain shape features from the current-derived features, and obtaining current-magnetic coordination node features from the current waveform timing features and electromagnetic time-domain shape features; obtaining linear displacement change features and vibration features of a moving contact in the high-voltage switch operating mechanism during operation from the mechanical feedback features, and obtaining vibration energy features from the current waveform timing features and vibration features; obtaining acceleration change features from the linear displacement change features, and obtaining electromagnetic driving force-movement acceleration coordination features from the acceleration change features and the electromagnetic time-domain shape features; fusing the current-magnetic coordination node features, the vibration energy features and the electromagnetic driving force-movement acceleration coordination features based on a lightweight Transformer model to obtain a fault risk assessment value; performing fault prediction on the high-voltage switch operating mechanism according to the fault risk assessment value to obtain a fault prediction result.

[0004] Preferably, the step of obtaining the current-magnetic coordination node feature according to the current waveform timing feature and the electromagnetic time-domain morphology feature comprises: obtaining a starting point time, a peak point time and an ending point time according to the current waveform timing feature; obtaining an electromagnetic action stage division point according to the starting point time, the peak point time and the ending point time, and obtaining a plurality of current feature values corresponding to the electromagnetic action stage division point; obtaining a current change gradient within a preset time according to the plurality of current feature values; obtaining a plurality of touch resistances within a preset time sequence according to the current waveform timing feature, obtaining a resistance influence factor according to the plurality of touch resistances, and obtaining a corrected current change gradient according to the resistance influence factor and the current change gradient; obtaining a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time-domain morphology feature, obtaining a plurality of continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to the plurality of current feature values, and obtaining an electromagnetic change gradient of adjacent time according to the plurality of continuous sampling electromagnetic force fluctuation amplitudes; dynamically associating the corrected current change gradient and the electromagnetic change gradient to obtain a current-magnetic dynamic association feature, and taking the current-magnetic dynamic association feature as the current-magnetic coordination node feature.

[0005] Preferably, the step of obtaining the vibration energy feature according to the current waveform timing feature and the vibration feature comprises: performing time-frequency transformation on the vibration feature to obtain a vibration frequency energy sequence, and performing collaborative decomposition on the vibration frequency energy sequence and the current waveform timing feature by using a multivariate empirical mode decomposition algorithm to obtain a plurality of sets of time-frequency aligned intrinsic mode function component sets, wherein the intrinsic mode function component sets comprise vibration energy intrinsic mode components and current waveform intrinsic mode components; calculating a plurality of vibration complex correlation coefficients and a plurality of current complex correlation coefficients of the vibration energy intrinsic mode components and the current waveform intrinsic mode components on different time scales, and integrating the plurality of vibration complex correlation coefficients and the plurality of current complex correlation coefficients by using a double-weight linear weighted fusion strategy to obtain a vibration-current fusion coefficient vector; generating a vibration energy timing feature by linear combination according to the vibration-current fusion coefficient vector and the vibration energy intrinsic mode components, and performing normalization processing on the vibration energy timing feature to obtain a standardized vibration energy sequence; extract a vibration peak point and a vibration duration from the vibration feature, and acquire a current rising edge time and a falling edge time according to the vibration peak point, the vibration duration and the current waveform timing feature, and acquire a vibration event synchronization identifier according to the current rising edge time and the falling edge time; based on the vibration event synchronization identifier and the standardized vibration energy sequence, calculate a vibration energy cumulative value by a weighted energy integration algorithm, and acquire a vibration energy distribution feature according to the vibration energy cumulative value and the vibration peak point; splicing the vibration energy distribution feature with the standardized vibration energy sequence to obtain a vibration energy feature.

[0006] As a preferred, the step of acquiring the electromagnetic driving force-motion acceleration coordination feature according to the acceleration change feature and the electromagnetic time domain morphology feature comprises: performing multi-scale decomposition on the linear displacement change feature to obtain a plurality of displacement intrinsic modal component sets, and acquiring a multi-scale acceleration component sequence according to the plurality of displacement intrinsic modal component sets; extract an electromagnetic force gradient sequence and an electromagnetic force fluctuation frequency from the electromagnetic time domain morphology feature, perform band self-adaptive decomposition on the electromagnetic force gradient sequence by using a variational modal decomposition algorithm to obtain an electromagnetic force intrinsic modal component set; perform time domain alignment processing on the multi-scale acceleration component sequence and the electromagnetic force intrinsic modal component set based on a dynamic time warping algorithm to obtain a phase synchronization coefficient and an amplitude coupling degree between components of each scale; construct a scale weight vector according to the phase synchronization coefficient and the amplitude coupling degree, and perform fusion on the multi-scale acceleration-electromagnetic force correlation feature by using an adaptive weighted fusion strategy to obtain a primary coordination feature sequence; acquire a motion stationarity index and a displacement overshoot according to the linear displacement change feature; acquire an electromagnetic force stable time and an overshoot amplitude according to the electromagnetic time domain morphology feature, and acquire a motion-electromagnetic coordination degree evaluation factor according to the electromagnetic force stable time, the overshoot amplitude, the motion stationarity index and the displacement overshoot; perform dynamic correction on the primary coordination feature sequence based on the motion-electromagnetic coordination degree evaluation factor to obtain a standardized coordination feature sequence; acquire an acceleration peak timing and an acceleration change rate extreme point according to the acceleration change feature, and acquire an acceleration-electromagnetic force dynamic correlation sequence according to the acceleration peak timing, the acceleration change rate extreme point and the electromagnetic force fluctuation frequency; and perform tensor splicing on the standardized coordination feature sequence and the acceleration-electromagnetic force dynamic correlation sequence to obtain an electromagnetic driving force-motion acceleration coordination feature.

[0007] As preferred, the step of fusing the current-electromagnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature based on the lightweight Transformer model to obtain the fault risk assessment value comprises: normalizing the current-electromagnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature to obtain a current-electromagnetic coordination node feature vector, a vibration energy feature vector and an electromagnetic driving force-motion acceleration collaborative feature vector; inputting the current-electromagnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector into a pre-trained lightweight Transformer model, wherein the lightweight Transformer model comprises an input embedding layer and a multi-layer encoder; mapping and merging the current-electromagnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector into a high-dimensional splicing feature vector based on the input embedding layer, and obtaining a corresponding vector module length according to the high-dimensional splicing feature vector, and taking the vector module length as the fault risk assessment value.

[0008] As preferred, the step of predicting the fault of the high-voltage switch operating mechanism according to the fault risk assessment value to obtain a fault prediction result comprises: obtaining historical normal operation data of the high-voltage switch operating mechanism, processing the historical normal operation data based on a lightweight Transformer model to generate a historical fault risk assessment value sequence, and constructing a health benchmark library according to the historical fault risk assessment value sequence, wherein the health benchmark library comprises a health benchmark interval and a health benchmark probability distribution; continuously monitoring the fault risk assessment value to obtain a risk assessment value sequence in a current time window; obtaining a fault decision threshold based on the health benchmark interval and the health benchmark probability distribution, and determining whether the risk assessment value sequence exceeds the fault decision threshold; if the risk assessment value sequence exceeds the fault decision threshold, it is determined that the high-voltage switch operating mechanism has a fault risk, and the fault risk is taken as the fault prediction result.

[0009] The application also provides a high-voltage switch operating mechanism fault prediction system, comprising: a first obtaining module configured to obtain operation data of a high-voltage switch operating mechanism in a preset time period, wherein the operation data comprises current-derived features and mechanical feedback features; a second obtaining module, configured to obtain a current waveform timing feature and an electromagnetic time-domain shape feature according to the current characteristic, and obtain a current-magnetic coordination node feature according to the current waveform timing feature and the electromagnetic time-domain shape feature; a third obtaining module, configured to obtain a linear displacement change feature and a vibration feature of a moving contact in the high-voltage switch operating mechanism in an operation process according to the mechanical feedback feature, and obtain a vibration energy feature according to the current waveform timing feature and the vibration feature; a fourth obtaining module, configured to obtain an acceleration change feature according to the linear displacement change feature, and obtain an electromagnetic driving force-motion acceleration coordination feature according to the acceleration change feature and the electromagnetic time-domain shape feature; a fusion module, configured to fuse the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration coordination feature based on a lightweight Transformer model to obtain a fault risk evaluation value; a prediction module, configured to perform fault prediction on the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result.

[0010] Preferably, the second obtaining module comprises: a first obtaining unit, configured to obtain a starting point time, a peak point time and an ending point time according to the current waveform timing feature; obtain an electromagnetic action stage division point according to the starting point time, the peak point time and the ending point time, and obtain a plurality of current feature values corresponding to the electromagnetic action stage division point; obtain a current change gradient within a preset time according to the plurality of current feature values; obtain a plurality of contact resistances within a preset time sequence according to the current waveform timing feature, obtain a resistance influence factor according to the plurality of contact resistances, and obtain a corrected current change gradient according to the resistance influence factor and the current change gradient; obtain a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time-domain shape feature, obtain a plurality of continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to the plurality of current feature values, and obtain an electromagnetic change gradient of adjacent time according to the plurality of continuous sampling electromagnetic force fluctuation amplitudes; dynamically associate the corrected current change gradient and the electromagnetic change gradient to obtain a current-magnetic dynamic association feature, and take the current-magnetic dynamic association feature as the current-magnetic coordination node feature.

[0011] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0012] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0013] The beneficial effects of this application are as follows: This invention achieves early and accurate prediction of faults in high-voltage switch operating mechanisms by extracting and fusing multiple features and combining them with a lightweight Transformer model. Compared with traditional single current signal analysis methods, this invention fully considers the coupling effect of electromagnetic and mechanical states. It captures electromagnetic anomalies through current-electromagnetic coordination node features, identifies mechanical vibration faults through vibration energy features, and detects motion incoordination through electromagnetic driving force-motion acceleration coordination features. This comprehensively covers the possible fault modes of the mechanism. The efficient fusion of the lightweight Transformer model further improves the accuracy and real-time performance of the assessment, effectively reduces the early fault false alarm rate, and provides a reliable guarantee for the safe operation of the power system. Attached Figure Description

[0014] Fig. 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0015] Fig. 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0016] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] like Figs. 1-2 As shown, this application provides a method for predicting faults in a high-voltage switch operating mechanism, comprising: S1. Obtain the operating data of the high-voltage switch operating mechanism within a preset time period, wherein the operating data includes current derivation characteristics and mechanical feedback characteristics; S2. Obtain the current waveform timing characteristics and electromagnetic time domain morphological characteristics based on the current derivation characteristics, and obtain the current-electromagnetic coordination node characteristics based on the current waveform timing characteristics and electromagnetic time domain morphological characteristics. S3. Obtain the linear displacement change characteristics and vibration characteristics of the moving contact in the high-voltage switch operating mechanism during operation based on the mechanical feedback characteristics, and obtain the vibration energy characteristics based on the current waveform timing characteristics and vibration characteristics. S4. Obtain acceleration change characteristics based on the linear displacement change characteristics, and obtain electromagnetic driving force-motion acceleration coordinated characteristics based on the acceleration change characteristics and the electromagnetic time domain morphological characteristics; S5, fusing the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature based on the lightweight Transformer model to obtain a fault risk evaluation value; S6, performing fault prediction on the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result.

[0019] As described in steps S1-S6, since the electromagnetic characteristics and mechanical state of the high-voltage switch operating mechanism are closely coupled in actual operation, when the mechanism transmission chain appears slight jamming, the current waveform changes weakly and is difficult to detect, but mechanical vibration and motion displacement have obvious abnormalities. The existing fault prediction usually uses single current signal analysis, which can lead to a high early fault missed report rate. The present application first obtains the operation data of the high-voltage switch operating mechanism in a preset time period, which includes current derived features and mechanical feedback features. The current derived features are usually collected by a current sensor to obtain the current signal of the operating coil, and the current waveform, electromagnetic force parameters, etc. are derived. The mechanical feedback features are obtained by displacement sensors and vibration sensors to obtain the linear displacement change and vibration signal of the moving contact. This step is the data basis, which ensures the comprehensiveness of the subsequent feature extraction and reflects the real-time state of the electromagnetic and mechanical systems of the mechanism in operation. According to the current derived features, current waveform time sequence features and electromagnetic time domain form features are obtained, and current-magnetic coordination node features are obtained according to the current waveform time sequence features and electromagnetic time domain form features. The current waveform time sequence features are the description of the change law of the current signal with time, including the key time nodes such as the starting point time, the peak point time, the ending point time and the change trend of the current in each time period, which can be obtained by time sequence analysis of the current derived features. The electromagnetic time domain form features reflect the change form of the electromagnetic force with time, including the electromagnetic force fluctuation amplitude, the electromagnetic force stable time and other parameters, which are obtained by electromagnetic force conversion and time domain analysis of the current derived features. By mining the collaborative correlation features of current and electromagnetic force, the electromagnetic action state of the high-voltage switch operating mechanism is accurately described. When the electromagnetic coil of the mechanism appears slight aging, the collaborative relationship between the current change gradient and the electromagnetic change gradient will shift slightly. This change can be captured in time by the feature extraction of this step, which provides a basis for early fault diagnosis. The technical effect lies in improving the description accuracy of the electromagnetic system operation state, and providing high-quality electromagnetic side feature support for subsequent comprehensive evaluation of mechanical features. According to the mechanical feedback feature, the linear displacement change feature and the vibration feature of the moving contact in the operating process of the high-voltage switch operating mechanism are obtained, and according to the current waveform time sequence feature and the vibration feature, the vibration energy feature is obtained, wherein the linear displacement change feature is obtained by preprocessing and analyzing the displacement data in the mechanical feedback feature, reflecting the movement trajectory and displacement change law of the moving contact in the operating process, and the vibration feature is the direct extraction of the vibration data in the mechanical feedback feature, including the basic parameters such as vibration amplitude, frequency and phase. When the vibration energy feature is obtained, the time sequence correlation between mechanical vibration and current signal is deeply mined to realize accurate extraction of the mechanical vibration energy feature. When the mechanism connecting rod appears slight jamming, the distribution characteristics of vibration energy will change obviously, and this change has a specific correlation with the time sequence characteristics of the current waveform. Through the analysis of this step, this fault signal can be accurately captured, and the technical effect lies in fully mining the cooperative relationship between mechanical vibration and electromagnetic signal, and improving the capture ability of early fault features of mechanical system. According to the linear displacement change feature, the acceleration change feature is obtained, and according to the acceleration change feature and the electromagnetic time domain form feature, the electromagnetic driving force-motion acceleration cooperative feature is obtained. The acceleration change feature is calculated by twice differentiating the linear displacement change feature, directly reflecting the acceleration change law of the moving contact motion, and can effectively reflect the smoothness of the mechanism mechanical motion. Through the cooperative relationship between the electromagnetic driving force and the mechanical motion acceleration, the electromagnetic-mechanical coupling state of the mechanism can be comprehensively evaluated. When the electromagnetic driving force of the mechanism electromagnet is insufficient, the cooperative relationship between the electromagnetic driving force and the motion acceleration will deviate obviously. Through the analysis of the dynamic response of the electromagnetic driving force and the mechanical motion acceleration, if the electromagnetic force fluctuation and the acceleration change are not coordinated during the acceleration process of the mechanism, it may cause transmission chain jamming or lubrication failure. Through multi-scale alignment and coordination degree evaluation, the precision of motion abnormality detection is improved. Secondly, it can also provide more comprehensive feature support for fault risk assessment. The current-electromagnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature are fused based on a lightweight Transformer model to obtain a fault risk evaluation value. First, the three types of features are normalized respectively to convert them into standardized feature vectors, eliminating the influence of different feature dimensions and numerical ranges. Then, the three types of feature vectors are input into a pre-trained lightweight Transformer model. The model includes an input embedding layer and a multi-layer encoder. The input embedding layer maps and combines the three types of feature vectors into a high-dimensional spliced feature vector. The multi-layer encoder performs deep processing on the high-dimensional spliced feature vector through a self-attention mechanism, fully excavating the internal correlation between the features. Finally, the fault risk evaluation value is obtained according to the vector length of the high-dimensional spliced feature vector. The lightweight Transformer model reduces the computational requirement while ensuring the accuracy of feature fusion by simplifying the model structure and optimizing the calculation method of the attention mechanism, realizing efficient fusion of multi-dimensional features and quantitative evaluation of fault risk. In addition, the self-attention mechanism of the model focuses on feature information that plays a key role in fault prediction, providing a basis for the efficiency of feature fusion and the accuracy of fault risk evaluation. According to the fault risk evaluation value, the fault of the high-voltage switch operating mechanism is predicted to obtain a fault prediction result. To ensure the reliability and accuracy of fault prediction, the current operating state needs to be compared and evaluated based on historical health data. In addition, dynamic evaluation of the mechanism state can avoid false positives and false negatives.

[0020] In one embodiment, the step S2 of obtaining the current-electromagnetic coordination node feature based on the current waveform timing feature and the electromagnetic time domain shape feature includes: S201, obtaining the starting point time, the peak point time and the ending point time based on the current waveform timing feature; S202, obtaining the electromagnetic action stage division point based on the starting point time, the peak point time and the ending point time, and obtaining a plurality of current feature values corresponding to the electromagnetic action stage division point; S203, obtaining the current change gradient within a preset time based on the plurality of current feature values; S204, obtaining a plurality of touch resistances within a preset time sequence based on the current waveform timing feature, obtaining a resistance influence factor based on the plurality of touch resistances, and obtaining a corrected current change gradient based on the resistance influence factor and the current change gradient; S205, obtain a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time domain mode characteristics, and obtain a plurality of continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to a plurality of current characteristic values, and obtain an electromagnetic change gradient of adjacent time according to a plurality of the continuous sampling electromagnetic force fluctuation amplitudes; S206, dynamically associate the corrected current change gradient and the electromagnetic change gradient to obtain a current-electromagnetic dynamic association feature, and take the current-electromagnetic dynamic association feature as a current-electromagnetic coordination node feature.

[0021] As described in the above steps S201-S206, the application first obtains a starting point time, a peak point time and an ending point time according to the current waveform time sequence characteristics, wherein the key time points are determined by detecting the rising edge, the extreme point and the falling edge of the current signal, the starting point corresponds to the time when the electromagnet starts to attract, the peak point corresponds to the time when the current maximum value, and the ending point corresponds to the time when the action is completed. The accurate extraction of these time points provides a time sequence reference for the subsequent stage division, so that the analysis can focus on the key electromagnetic state transition nodes in the mechanism action process; According to the starting point time, the peak point time and the ending point time, obtain an electromagnetic action stage division point, and obtain a plurality of corresponding current characteristic values according to the electromagnetic action stage division point, wherein the electromagnetic action stage usually includes a starting stage, an acceleration stage and a stable stage, each stage corresponds to a different current change mode, in the starting stage, the current rises rapidly, in the acceleration stage, the current may fluctuate, and in the stable stage, the current tends to be stable, and by extracting the current characteristic values in stages, such as the average current, the peak current or the current integral value of each stage, the working state of the electromagnetic system in different action stages can be described more finely, and a data basis is provided for subsequent gradient analysis; According to the plurality of current characteristic values, a current change gradient in a preset time is obtained, the preset time being determined according to the time length of each stage of electromagnetic action, for example, the time length of the electromagnetic force rising stage is 0.05 seconds, and the preset time can be set as the time length of the stage. The plurality of current characteristic values obtained in the stage are arranged in time sequence, the difference between the current characteristic values of two adjacent time points is calculated, and the difference is divided by the time interval of the adjacent time points to obtain the current change rate in each time interval. Then, the average of the current change rates in all time intervals is taken, and the current change gradient in the preset time is obtained. The physical meaning of this step is to quantify the change rate of the current in different electromagnetic action stages. The size of the current change gradient directly reflects the speed of the current increase or decrease in the electromagnetic coil, and the speed of the current change is related to the change rate of the electromagnetic force. For example, when the current change gradient is large, it indicates that the electromagnetic force rises or falls quickly. If it exceeds the normal range, it may mean that there is a short circuit risk in the coil. This step converts the dynamic change of the current into a quantifiable index by calculating the current change gradient. The technical effect is to provide a quantifiable current dynamic index for subsequent analysis of the correlation between the current and the electromagnetic force, which facilitates more accurate judgment of whether the current change is normal. S204, according to the current waveform timing characteristics, a plurality of contact resistances in a preset timing are acquired, and an electrical resistance influence factor is acquired according to the plurality of contact resistances, and a corrected current change gradient is acquired according to the electrical resistance influence factor and the current change gradient, wherein the contact resistance is the contact resistance of the moving contact of the high-voltage switch operating mechanism and the static contact when the moving contact is in contact with the static contact, and the size of the contact resistance will affect the actual change of the current. The contact resistance at a plurality of time points is calculated according to Ohm's law (resistance = voltage / current) by collecting the voltage drop data of the contact in the preset timing and combining the current characteristic value in the timing. When the electrical resistance influence factor is acquired, the average value of the plurality of contact resistances in the preset timing is calculated, and then the average value is compared with the standard contact resistance value to obtain the contact resistance deviation rate. After the contact resistance deviation rate is normalized, the electrical resistance influence factor can be obtained. The min-max normalization method is used for normalization processing, and the deviation rate is mapped to 0-1. The greater the deviation rate, the closer the electrical resistance influence factor is to 1, and vice versa. When the contact resistance is corrected, the formula "corrected current change gradient = current change gradient x (1-electrical resistance influence factor)" is used for calculation. When the contact resistance is too large, the electrical resistance influence factor increases, and the corrected current change gradient decreases accordingly, so as to eliminate the influence of the large contact resistance on the calculation result of the current change gradient. The step is to eliminate the interference of the contact resistance change on the current change gradient. Secondly, since the contact resistance will change with the factors such as contact wear and oxidation, and further affect the actual change of the current, if the correction is not performed, the calculated current change gradient will not accurately reflect the current change of the coil itself. At the same time, by introducing the electrical resistance influence factor to correct the current change gradient, the corrected gradient is closer to the real current change state of the coil, and the technical effect is to improve the accuracy of the current change gradient, and to lay a reliable foundation for the subsequent correlation with the electromagnetic change gradient; S205, obtain a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time domain shape characteristics, and obtain electromagnetic change gradients of adjacent time points according to a plurality of the continuous sampling electromagnetic force fluctuation amplitudes, wherein the electromagnetic time domain shape characteristics are derived through an electromagnetic force calculation formula, according to the Ampere force formula, the electromagnetic force is proportional to the square of the current, combined with the number of turns of the electromagnetic coil of the mechanism, the cross-sectional area of the core and other parameters, the current characteristic value can be converted into the corresponding electromagnetic force value, and the electromagnetic time domain shape characteristics can be obtained after being arranged in time sequence. When a plurality of electromagnetic force fluctuation amplitudes are obtained, the electromagnetic force values in the electromagnetic time domain shape characteristics are analyzed, the difference between the electromagnetic force value at each time and the average value of the electromagnetic force at the stage of the time is calculated, and the absolute value of the difference is the electromagnetic force fluctuation amplitude at the time. The preset time is consistent with the foregoing, and then the electromagnetic force fluctuation amplitude is continuously collected at a fixed time interval within the preset time to obtain a plurality of continuous sampling electromagnetic force fluctuation amplitudes, and the electromagnetic change gradients of adjacent time points are calculated. The method is similar to the current change gradient, the difference between the adjacent two continuous sampling electromagnetic force fluctuation amplitudes is calculated, divided by the time interval of adjacent time points, the electromagnetic change rate in each time interval is obtained, and the average value of the electromagnetic change rate in all time intervals is obtained, that is, the electromagnetic change gradient of adjacent time points. This can quantify the change rate of the electromagnetic force fluctuation, the electromagnetic change gradient reflects the increase and decrease speed of the electromagnetic force fluctuation, and the electromagnetic change gradient and the current change gradient jointly constitute the core index of the dynamic change of the electromagnetic system. For example, when the electromagnetic change gradient and the current change gradient do not match obviously, it may mean that there is an abnormality in the electromagnetic force transmission link. This step calculates the electromagnetic change gradient to provide a quantitative electromagnetic dynamic index for subsequent correlation with the current change gradient; S206, dynamically correlate the corrected current change gradient and the electromagnetic change gradient to obtain a current-electromagnetic dynamic correlation characteristic, and take the current-electromagnetic dynamic correlation characteristic as a current-electromagnetic coordination node characteristic. The corrected current change gradient and the electromagnetic change gradient are normalized to obtain a corrected current change gradient normalized value and an electromagnetic change gradient normalized value, wherein the current-electromagnetic coordination node characteristic = corrected current change gradient normalized value * corrected current change gradient weight value + electromagnetic change gradient normalized value * (1-electromagnetic change gradient normalized value). In this way, the dynamic changes of the current and the electromagnetic force can be correlated to form a comprehensive characteristic that can comprehensively reflect the cooperative relationship between the two. This characteristic directly reflects the running state of the electromagnetic system. For example, when the correlation coefficient is lower than the normal threshold, it may mean that the coil is aging, causing the current change to be unable to effectively convert into electromagnetic force change, or there is mechanical jamming, causing the electromagnetic force to be unable to be normally transmitted. This step constructs the coordination node characteristic through dynamic correlation to provide a precise electromagnetic side cooperative characteristic for subsequent fusion of mechanical characteristics for fault risk assessment.

[0022] In one embodiment, the step S3 of obtaining the vibration energy feature according to the current waveform time sequence feature and the vibration feature comprises: S301, performing time-frequency transformation on the vibration feature to obtain a vibration frequency domain energy sequence, and performing cooperative decomposition on the vibration frequency domain energy sequence and the current waveform time sequence feature by using a multivariate empirical mode decomposition algorithm to obtain a plurality of sets of time-frequency aligned intrinsic mode function component sets, wherein the intrinsic mode function component sets include vibration energy intrinsic mode components and current waveform intrinsic mode components; S302, calculating a plurality of vibration complex correlation coefficients and a plurality of current complex correlation coefficients of the vibration energy intrinsic mode components and the current waveform intrinsic mode components on different time scales, and integrating the plurality of vibration complex correlation coefficients and the plurality of current complex correlation coefficients by using a double-weight linear weighting fusion strategy to obtain a vibration-current fusion coefficient vector; S303, generating a vibration energy time sequence feature by linear combination according to the vibration-current fusion coefficient vector and the vibration energy intrinsic mode components, and performing normalization processing on the vibration energy time sequence feature to obtain a standardized vibration energy sequence; S304, extracting a vibration peak point and a vibration duration from the vibration feature, and obtaining a current rising edge time and a falling edge time according to the vibration peak point, the vibration duration and the current waveform time sequence feature, and obtaining a vibration event synchronization identifier according to the current rising edge time and the falling edge time; S305, calculating a vibration energy cumulative value by using a weighted energy integration algorithm based on the vibration event synchronization identifier and the standardized vibration energy sequence, and obtaining a vibration energy distribution feature according to the vibration energy cumulative value and the vibration peak point; S306, splicing the vibration energy distribution feature and the standardized vibration energy sequence to obtain a vibration energy feature.

[0023] As described in steps S301-S306 above, the application first performs time-frequency transformation on the original vibration features collected by the vibration sensor (such as an accelerometer), using short-time Fourier transform or wavelet transform, to convert the time-domain vibration signal into a vibration frequency energy sequence, thereby revealing the distribution of vibration energy in the frequency dimension. Subsequently, the vibration frequency energy sequence and the current waveform time series features extracted from the current sensor signal are used as two input channels for collaborative decomposition using the multivariate empirical mode decomposition algorithm. The MEMD algorithm can adaptively decompose the two signals into multiple sets of time-frequency aligned intrinsic mode function components, including vibration energy intrinsic mode components and current waveform intrinsic mode components. It can align different physical sources but time-correlated signals in the time-frequency domain to the same internal fluctuation mode, laying the foundation for subsequent analysis of their correlation in the common time-frequency scale of the two signals. For example, it can decompose the high-frequency vibration components generated by mechanical impact during the closing process of the mechanism into the same intrinsic mode as the corresponding fluctuation components in the current waveform generated by the sudden change of electromagnetic force, thereby revealing the coupling relationship between the two in a specific time scale and providing mechanical side information for subsequent fusion with current-electromagnetic coordination node features. Based on the vibration energy intrinsic mode components and the current waveform intrinsic mode components, the vibration energy and the current waveform are calculated to obtain multiple vibration complex correlation coefficients and multiple current complex correlation coefficients in different time scales, and the multiple vibration complex correlation coefficients and the multiple current complex correlation coefficients are integrated by a double-weight linear weighted fusion strategy to obtain a vibration-current fusion coefficient vector. Based on the vibration energy intrinsic mode components and the current waveform intrinsic mode components obtained by decomposition, the complex correlation coefficients of the two in different time scales (i.e. different IMF components) are calculated. The complex correlation coefficient here is used to measure the linear correlation degree between the two sets of intrinsic mode component sequences, which takes into account the amplitude and phase information. For each decomposition time scale, a vibration complex correlation coefficient and a current complex correlation coefficient are calculated respectively to quantify the correlation strength of the vibration energy and the current waveform. Subsequently, these coefficients are integrated by a double-weight linear weighted fusion strategy. This strategy can assign two weights to each correlation coefficient: one is based on the energy proportion of the IMF component (objective weight), and the other is based on the sensitivity of the component to the fault in prior knowledge (historical experience data) (subjective weight). Finally, the weighted sum is obtained to obtain a vibration-current fusion coefficient vector, and the calculation formula of the double-weight linear weighted fusion is: ; Wherein, represents the comprehensive complex correlation coefficient, represents the vibration complex correlation coefficient, represents the weight value of the vibration complex correlation coefficient (the weight size is obtained based on the energy proportion of the IMF component), represents the vibration fault sensitivity coefficient, represents a current fault sensitivity coefficient, represents a current complex correlation coefficient, so that a plurality of said vibration complex correlation coefficients and a plurality of said current complex correlation coefficients are calculated according to the above calculation formula to obtain a plurality of comprehensive complex correlation coefficients, and the comprehensive complex correlation coefficients and a plurality of comprehensive complex correlation coefficients are combined according to a preset time sequence to obtain a vibration-current fusion coefficient vector. The technical effect of this step is to abandon the practice of equal observation of all frequency components, and instead focus on those key time scales that best reflect the state of mechanical-electrical coupling according to their physical importance and statistical significance, thereby improving the relevance of the features; According to the vibration-current fusion coefficient vector and the vibration energy intrinsic modal component, a vibration energy time sequence feature is calculated by linear combination, and the vibration energy time sequence feature is normalized to obtain a standardized vibration energy sequence. Each vibration energy IMF component is multiplied by the corresponding weight in the fusion coefficient vector, and then all the weighted components are reconstructed to generate a new vibration energy time sequence feature. This is a vibration energy signal weighted by current correlation, which amplifies the vibration components with strong synchronization with the electromagnetic driving process and suppresses irrelevant random vibration noise. Then, the time sequence feature is normalized to obtain a standardized vibration energy sequence. This makes the vibration energy of different operations and different devices comparable, providing convenience for subsequent threshold judgment and model analysis; From the vibration feature, a vibration peak point and a vibration duration are extracted, and a current rising edge time and a falling edge time are obtained according to the vibration peak point, the vibration duration and the current waveform time sequence feature. These time points correspond to the start and end of the electromagnet action, and a vibration event synchronization identifier is obtained according to the current rising edge time and the falling edge time. The identifier can be a quantitative indicator for identifying whether a specific mechanical vibration event (such as contact closure impact) is accurately synchronized in time with the expected electromagnetic driving action. For example, if the vibration peak point is significantly later than the current rising edge time, it indicates that the transmission mechanism is stuck. This step introduces the macro event time sequence relationship into the feature, so that the feature can capture the delay or advance phenomenon of mechanical response to electromagnetic instructions; Based on the vibration event synchronization identifier and the standardized vibration energy sequence, a vibration energy accumulation value is calculated by a weighted energy integration algorithm, wherein the specific calculation process is as follows: the vibration event synchronization identifier refers to two forms and corresponds to two strategies, a Boolean type, for each detected vibration event (usually centered on a vibration peak point), the identifier is a binary quantity, for example: 1 indicates that the vibration event is synchronized in time with the current action (such as rising edge / falling edge), and 0 indicates that it is not synchronized, a continuous type, the identifier is a continuous value between 0 and 1, used to quantify the degree of synchronization. For example, it can be based on the time difference between the vibration peak point and the current key timing point (such as the rising edge time), wherein for each identified vibration event (vibration peak point), according to the vibration duration of the event, an integral time window centered on the peak point or starting from the peak point is defined, and finally, the standardized vibration energy sequence is integrated within the time window corresponding to each vibration event, and the result is multiplied by the weight of the event. Sum all events to get the total vibration energy accumulation value, this algorithm will give higher weight to the vibration events with good synchronization with current action, and integrate the corresponding segment of the standardized vibration energy sequence, so as to calculate the vibration energy that can better represent the effective mechanical action. Then, combined with the vibration peak point (reflecting the impact intensity), the vibration energy distribution characteristics such as energy concentration and peak energy ratio are extracted from the relationship between the energy accumulation value and the peak point. These distribution characteristics can distinguish different types of mechanical states, such as a severe impact and a sustained vibration, whose energy distribution patterns are different. The vibration energy distribution characteristics and the standardized vibration energy sequence are spliced to obtain vibration energy characteristics, which contain the original energy sequence information of vibration (standardized and current correlation weighted) and its macro statistical distribution and event synchronization information, and constitute a high-dimensional feature that describes the mechanical state very comprehensively.

[0024] In one embodiment, the step S4 of acquiring the electromagnetic driving force-motion acceleration coordination feature according to the acceleration change feature and the electromagnetic time-domain morphology feature comprises: S401, performing multi-scale decomposition on the linear displacement change feature to obtain a plurality of displacement intrinsic modal component sets, and acquiring a multi-scale acceleration component sequence according to the plurality of displacement intrinsic modal component sets; S402, extracting an electromagnetic force gradient sequence and an electromagnetic force fluctuation frequency from the electromagnetic time-domain morphology feature, and performing frequency band adaptive decomposition on the electromagnetic force gradient sequence by using a variational modal decomposition algorithm to obtain an electromagnetic force intrinsic modal component set; S403, time domain alignment processing is performed on the multi-scale acceleration component sequence and the set of electromagnetic force intrinsic modal components based on a dynamic time warping algorithm, to obtain phase synchronization coefficients and amplitude coupling degrees between components of each scale; S404, a scale weight vector is constructed according to the phase synchronization coefficients and the amplitude coupling degrees, and a self-adaptive weighted fusion strategy is used to fuse multi-scale acceleration-electromagnetic force correlation characteristics, to obtain a primary collaborative feature sequence; S405, a motion stability index and a displacement overshoot are obtained according to the linear displacement change characteristics; S406, an electromagnetic force stable time and an overshoot amplitude are obtained according to the electromagnetic time domain shape characteristics, and a motion-electromagnetic coordination degree evaluation factor is obtained according to the electromagnetic force stable time, the overshoot amplitude, the motion stability index and the displacement overshoot; S407, the primary collaborative feature sequence is dynamically modified based on the motion-electromagnetic coordination degree evaluation factor, to obtain a standardized collaborative feature sequence; S408, an acceleration peak timing and an acceleration change rate extreme point are obtained according to the acceleration change characteristics, and an acceleration-electromagnetic force dynamic correlation sequence is obtained according to the acceleration peak timing, the acceleration change rate extreme point and the electromagnetic force fluctuation frequency; the standardized collaborative feature sequence and the acceleration-electromagnetic force dynamic correlation sequence are tensor spliced, to obtain an electromagnetic driving force-motion acceleration collaborative feature.

[0025] As described above in steps S401-S408, the linear displacement change characteristics collected by a linear displacement sensor (such as an LVDT) are subjected to multi-scale decomposition. This decomposition can use algorithms such as ensemble empirical mode decomposition or wavelet packet transform to decompose the original displacement-time sequence into a set of displacement intrinsic modal components of different frequency components. These components respectively represent the dynamic characteristics of different frequency bands in the motion process of the mechanism, for example, low-frequency components correspond to the motion trend of the main body, and high-frequency components correspond to small shocks or vibrations of the transmission chain. Then, the displacement intrinsic modal components are subjected to secondary differentiation (numerical calculation can use central difference method), i.e. a multi-scale acceleration component sequence is obtained, which decouples the overall motion acceleration to different time scales, and can completely retain the acceleration characteristics of the original motion in each frequency band, so that subsequent fine analysis of the correlation between mechanical dynamics of different frequency bands and specific electromagnetic force fluctuation components can be performed; The electromagnetic force gradient sequence and the electromagnetic force fluctuation frequency are extracted from the electromagnetic time-domain shape features. The electromagnetic force gradient sequence is calculated by first-order difference calculation on the electromagnetic force-time sequence, reflecting the speed of electromagnetic force change; the electromagnetic force fluctuation frequency can be obtained by frequency spectrum analysis or zero-crossing detection. Then, the electromagnetic force gradient sequence is decomposed by the VMD (Variational Mode Decomposition) algorithm, which can robustly decompose the input signal into a set of intrinsic mode components with central frequency by constructing and solving the variational problem. Compared with EMD (Empirical Mode Decomposition), VMD can effectively avoid modal aliasing, so that the decomposed components are more distinguishable in the frequency domain, thus more clearly separating different fluctuation modes in the electromagnetic force change; The multi-scale acceleration component sequence and the electromagnetic force intrinsic mode component set are aligned in time domain based on the dynamic time warping algorithm. Due to the inherent delay between electromagnetic response and mechanical motion, and the possible slight asynchronization of signal acquisition path, direct point-to-point comparison will produce errors. The DTW algorithm finds the optimal nonlinear bending path between the two sequences to achieve the best alignment on the time axis. On the basis of alignment, the phase synchronization coefficient (such as calculating the cosine value of the instantaneous phase difference of the aligned sequence) and the amplitude coupling degree (such as calculating the Pearson correlation coefficient of the aligned sequence) between the components at different scales are calculated. The technical effect of this step is that it accurately quantifies the matching degree of mechanical acceleration and electromagnetic force gradient in time and amplitude at a specific frequency component, providing a scientific basis for subsequent weighted fusion; According to the phase synchronization coefficient and the amplitude coupling degree between the components at different scales, a scale weight vector is constructed. The assignment principle of the weight vector is that the scale components with good phase synchronization and high amplitude coupling degree are given greater weight, indicating that the electromechanical coupling relationship of this scale is close and the information is reliable. Then, an adaptive weighted fusion strategy is used to linearly combine the acceleration-electromagnetic force correlation features at different scales according to the weight vector to obtain a primary cooperative feature sequence. This step realizes the first information purification and strengthens the contribution of scale components that can provide clear and stable electromechanical coupling evidence. The primary cooperative feature obtained finally is more focused and reliable; From the original linear displacement change feature, the motion stability index (such as the variance of the acceleration derivative) and the displacement overshoot (the difference between the actual stopping position and the target position) are further extracted, and from the electromagnetic time-domain shape features, the electromagnetic force settling time (the time from the start of the action to the force value stabilizing within the allowed error band) and the overshoot amplitude (the amount by which the electromagnetic force peak value exceeds the steady-state value) are extracted. The electromagnetic force settling time, overshoot amplitude, motion stability index and displacement overshoot are fused by weighted calculation to obtain a motion-electromagnetic coordination degree evaluation factor, and the calculation formula is: ; wherein, This represents the evaluation factor for motion-electromagnetic coordination. Indicates the settling time of the electromagnetic force. This represents the reference value for the electromagnetic force stabilization time under normal conditions. express The weight value, Indicates the overshoot amplitude. This represents the reference value for the electromagnetic force overshoot amplitude under normal conditions. express The weight value, Indicators representing motion stability This represents the baseline value for motion stability under normal conditions. express The weight value, Indicates displacement overshoot. This represents the reference value for displacement overshoot under normal conditions. express The weighting factor quantifies the overall matching degree between electromagnetic drive and mechanical response. For example, when the coil ages, the electromagnetic force rises more slowly (increases), requiring a larger driving force to complete the same movement, causing the mechanical system to bear additional stress. In this case, the matching factor will decrease significantly, and even if individual parameters are still within the allowable range, it can provide an early warning of potential faults. The primary cooperative feature sequence is dynamically corrected based on a nonlinear correction function to obtain a standardized cooperative feature sequence. The formula for the nonlinear correction function is as follows: ; in, Represents a standardized collaborative feature sequence. This represents the primary cooperative feature sequence, and k represents the adjustment parameter (a dimensionless constant that controls the sensitivity to nonlinear correction). A larger value indicates a higher sensitivity to nonlinear correction. The more sensitive the change, the smaller the value; the smoother the correction. Indicates the settling time of the electromagnetic force. This represents the reference value for the electromagnetic force stabilization time under normal conditions. This represents the motion-electromagnetic coordination evaluation factor, which allows the feature to not only reflect instantaneous correlations but also incorporate a macroscopic evaluation of the coordination of the entire operation process, thereby improving the robustness of the feature. According to the acceleration change characteristics obtained by the quadratic differentiation of the linear displacement, the acceleration peak time sequence and the acceleration change rate extreme point are extracted. In combination with the magnetic force fluctuation frequency, an acceleration-magnetic force dynamic correlation sequence is generated. The sequence can be a code containing the relative time sequence relationship between the key time points (such as the peak value, the extreme point) and the main fluctuation period of the electromagnetic force. Finally, the standardized collaborative feature sequence and the acceleration-magnetic force dynamic correlation sequence are spliced by tensor to form the final electromagnetic driving force-motion acceleration collaborative feature. This final feature contains continuous collaborative information after multi-scale analysis, time domain alignment and coordination correction, as well as discrete time sequence correlation information of key dynamic events.

[0026] In one embodiment, the step S5 of obtaining the fault risk evaluation value according to the light-weight Transformer model based on the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature comprises: S501, the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature are normalized to obtain a current-magnetic coordination node feature vector, a vibration energy feature vector and an electromagnetic driving force-motion acceleration collaborative feature vector; S502, the current-magnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector are input into a pre-trained light-weight Transformer model, wherein the light-weight Transformer model includes an input embedding layer and a multi-layer encoder; S503, the current-magnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector are mapped and combined into a high-dimensional splicing feature vector based on the input embedding layer, and the corresponding vector module length is obtained according to the high-dimensional splicing feature vector, and the vector module length is taken as the fault risk evaluation value.

[0027] As described in steps S501-S503 above, the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature are normalized to obtain a current-magnetic coordination node feature vector, a vibration energy feature vector and an electromagnetic driving force-motion acceleration collaborative feature vector. This can eliminate the differences in dimension and order of magnitude of different physical quantities (such as current gradient, vibration energy, acceleration), providing a basis for subsequent light-weight Transformer model training; The current-electromagnetic coordination node normalized feature vector, the vibration energy normalized feature vector, and the electromagnetic driving force-motion acceleration collaborative normalized feature vector are input into a pre-trained lightweight Transformer model optimized for the high-voltage switchgear monitoring scene, the core structure of which includes an input embedding layer and a multi-layer encoder. The role of the input embedding layer is to map the three feature vectors to a unified higher-dimensional feature space and combine them (for example, through a splicing operation) into a comprehensive high-dimensional spliced feature vector. This mapping process is usually implemented through a learnable linear transformation matrix, so that features from different sources have comparability in the same semantic space. Subsequently, this high-dimensional spliced feature vector is fed into the multi-layer encoder. Each layer of the encoder contains a multi-head self-attention mechanism and a feedforward neural network. The self-attention mechanism can automatically calculate the correlation weights between the elements (which can be understood as different feature dimensions or features at different time steps) of the feature vector, thereby capturing the deep dependence between current, vibration, motion acceleration and other physical quantities, regardless of whether the relationship is short-range or long-range. Lightweight design is reflected in: fewer layers of encoders (for example, 2-4 layers), moderate number of attention heads (for example, 4 heads), appropriate reduction of feedforward network hidden layer dimensions, and the use of techniques such as grouped convolution or knowledge distillation to further compress the model. This design retains the powerful feature correlation capabilities of the Transformer while significantly reducing the model's parameter size and computational complexity, enabling it to be deployed on edge devices with limited computing resources and meeting the real-time requirements of high-voltage switchgear state monitoring; Based on the high-dimensional spliced feature vector output by the input embedding layer, the corresponding vector length (i.e. Euclidean norm) is calculated, and this length is taken as the final fault risk assessment value. The calculation of the vector length is the square sum of all dimension values of the comprehensive feature vector, and then the square root is taken. Its physical meaning is that it quantifies the overall deviation of the current state feature from an implicit "healthy origin" in the common high-dimensional feature space. In the ideal healthy state, after a large amount of normal data training, the model expects the length of its comprehensive feature vector to remain in a relatively low and stable range. When an anomaly occurs in the mechanism, whether it is current-electromagnetic imbalance, vibration energy anomaly, or motion acceleration disorder, it will cause the comprehensive feature vector to deviate in multiple dimensions, resulting in an increase in its length. Therefore, as a scalar output, this length can intuitively and comprehensively reflect the overall health degradation of the device, with a higher value representing a greater risk of failure; In summary, the scheme provided by the application has the following advantages: the normalization processing ensures the fairness and stability of feature fusion; the lightweight Transformer model can automatically and efficiently learn the complex nonlinear correlation between different physical features through its powerful self-attention mechanism, avoiding the difficulties and limitations of manually designed fusion rules; and finally using the vector length as the evaluation value provides an unsupervised or semi-supervised fault risk measure with clear geometric interpretation and physical meaning, making the decision-making process more transparent and reliable. This method not only improves the recognition ability of composite faults and early weak faults, but also is more suitable for real-time online monitoring and prediction in embedded systems in industrial sites due to its lightweight characteristics.

[0028] In one embodiment, the step S6 of performing fault prediction on the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result comprises: S601, obtaining historical normal operation data of the high-voltage switch operating mechanism, processing the historical normal operation data based on a lightweight Transformer model to generate a historical fault risk evaluation value sequence, and constructing a health benchmark library according to the historical fault risk evaluation value sequence, wherein the health benchmark library comprises a health benchmark interval and a health benchmark probability distribution; S602, continuously monitoring the fault risk evaluation value to obtain a risk evaluation value sequence in a current time window; S603, obtaining a fault decision threshold based on the health benchmark interval and the health benchmark probability distribution, and determining whether the risk evaluation value sequence exceeds the fault decision threshold; If the risk evaluation value sequence exceeds the fault decision threshold, it is determined that the high-voltage switch operating mechanism has a fault risk, and the fault risk is taken as the fault prediction result.

[0029] As described in steps S601-S603 above, the application first obtains historical normal operation data of the high-voltage switch operating mechanism in the known normal state. These data are usually obtained from the trial operation stage after successful commissioning of the equipment, periodic preventive tests, or operation records confirmed to be normal in long-term monitoring. These historical data are input into the aforementioned lightweight Transformer model that has been trained to generate a series of historical fault risk assessment values, forming a sequence of historical fault risk assessment values representing the health status of the equipment. Based on this sequence, a health benchmark library is constructed through statistical methods. The health benchmark interval can usually be a certain percentile interval of the sequence, for example, an interval from the 5th percentile to the 95th percentile, indicating that the assessment value of the equipment in the healthy state has a high probability of falling within this range. The health benchmark probability distribution can be obtained by kernel density estimation or fitting a specific probability distribution (such as Gaussian distribution, Gamma distribution) to the sequence, which describes the likelihood of the assessment value in the healthy state. This provides a dynamic and personalized reference system for subsequent fault judgment, which captures the inherent fluctuation characteristics of the specific equipment under normal working conditions, making the judgment standard no longer rigid, but closely related to the running history of the equipment itself; The current equipment state is continuously monitored. It collects the fault risk assessment values calculated by the lightweight Transformer model in real time within a set current time window, forming a sequence of risk assessment values. This time window can include data from the last several operations, such as the last 10 operations or all operation records within the last 24 hours. Using a time window for sequence monitoring instead of single-point judgment has the advantage of smoothing accidental transient disturbances and capturing trend changes in the state. For example, a single operation's assessment value may be accidentally high due to noise, but if the sequence values within a window show a consistent upward trend or most points deviate from the normal range, it is more likely to indicate real performance degradation. This step ensures the stability of state assessment and sensitivity to trends; The current risk assessment value sequence is finally fault-decided based on the health benchmark library. First, a fault decision threshold is obtained based on the health benchmark interval and the health benchmark probability distribution. This threshold is not a single fixed value, but a dynamically calculated value. For example, the fault decision threshold can be set to the upper limit of the health benchmark interval (such as the 95th percentile), or it can be a critical value at a certain significance level calculated based on the health benchmark probability distribution (such as setting the 99th percentile of the probability distribution, or using the EVT to estimate the threshold of rare events in the health state). Then, it is determined whether the risk assessment value sequence in the current time window exceeds the fault decision threshold. The judgment logic here can be diverse, such as requiring a plurality of points in the sequence to exceed the threshold, or the mean / slide average of the sequence to exceed the threshold, or the current value to have a probability of less than a very low level (such as a p-value less than 0.01) in the health benchmark probability distribution. If the judgment condition is met, it is determined that there is a fault risk in the high-voltage switch operating mechanism, and this is output as the fault prediction result. This statistical distribution-based decision mechanism greatly reduces false positives caused by random fluctuations, and for slowly developing early faults, the evaluation value will gradually deviate from the high-density area of the health probability distribution, so it can be captured in time, effectively reducing the false negative rate; In summary, the above steps realize the upgrade of fault prediction from "absolute threshold judgment" to "relative historical state deviation judgment" by constructing an individualized health benchmark library and introducing a dynamic decision threshold based on statistical theory. Compared with the traditional fixed threshold method, the present application can adapt to individual differences of different devices and changes in operating conditions, making the warning threshold more reasonable and accurate. By continuously monitoring the sequence data and using statistical significance testing logic, this method enhances the robustness to noise and has higher detection sensitivity to gradual degradation of device performance. Finally, the reliability of the high-voltage switch operating mechanism fault risk prediction result is ensured, and a scientific and effective decision basis for predictive maintenance is provided.

[0030] The present application also provides a high-voltage switch operating mechanism fault prediction system, comprising: A first acquisition module for acquiring operating data of a high-voltage switch operating mechanism in a preset time period, the operating data including current-derived features and mechanical feedback features; A second acquisition module for acquiring current waveform timing features and electromagnetic time-domain shape features based on the current-derived features, and acquiring current-electromagnetic coordination node features based on the current waveform timing features and the electromagnetic time-domain shape features; A third acquisition module for acquiring linear displacement change features and vibration features of a moving contact in the high-voltage switch operating mechanism during operation based on the mechanical feedback features, and acquiring vibration energy features based on the current waveform timing features and the vibration features; a fourth obtaining module, configured to obtain an acceleration variation feature according to the linear displacement variation feature, and obtain an electromagnetic driving force-movement acceleration coordination feature according to the acceleration variation feature and the electromagnetic time-domain form feature; a fusion module, configured to fuse the current-electromagnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-movement acceleration coordination feature based on a lightweight Transformer model to obtain a fault risk evaluation value; a prediction module, configured to perform fault prediction on the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result.

[0031] In one embodiment, the second obtaining module comprises: a first obtaining unit, configured to obtain a starting point time, a peak point time and an ending point time according to the current waveform time sequence feature; obtain an electromagnetic action stage division point according to the starting point time, the peak point time and the ending point time, and obtain a plurality of current feature values corresponding to the electromagnetic action stage division point; obtain a current variation gradient within a preset time according to the plurality of current feature values; obtain a plurality of touch resistances within a preset time sequence according to the current waveform time sequence feature, obtain a resistance influence factor according to the plurality of touch resistances, and obtain a corrected current variation gradient according to the resistance influence factor and the current variation gradient; obtain a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time-domain form feature, obtain a plurality of continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to the plurality of current feature values, and obtain an electromagnetic variation gradient of adjacent time according to the plurality of continuous sampling electromagnetic force fluctuation amplitudes; dynamically associate the corrected current variation gradient and the electromagnetic variation gradient to obtain a current-electromagnetic dynamic association feature, and take the current-electromagnetic dynamic association feature as a current-electromagnetic coordination node feature.

[0032] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0033] The application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the above method.

[0034] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, databases, or other media in this application and in examples used herein, unless specifically stated otherwise, can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct Rambus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0035] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article, or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article, or method. Without more limitations, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article, or method that includes the element.

[0036] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, based on the content of the present application specification and drawings, is also included in the patent protection scope of the present application.

Claims

1. A high-voltage switch operating mechanism failure prediction method characterized by comprising: The method comprises the following steps: acquiring operation data of the high-voltage switch operating mechanism within a preset time period, wherein the operation data comprises current-derived features and mechanical feedback features; acquiring current waveform timing features and electromagnetic time-domain shape features according to the current-derived features, and acquiring current-magnetic coordination node features according to the current waveform timing features and the electromagnetic time-domain shape features; acquiring linear displacement change features and vibration features of a moving contact in the high-voltage switch operating mechanism during operation according to the mechanical feedback features, and acquiring vibration energy features according to the current waveform timing features and the vibration features; acquiring acceleration change features according to the linear displacement change features, and acquiring electromagnetic driving force-motion acceleration coordination features according to the acceleration change features and the electromagnetic time-domain shape features; fusing the current-magnetic coordination node features, the vibration energy features, and the electromagnetic driving force-motion acceleration coordination features based on a lightweight Transformer model to obtain a fault risk assessment value; performing fault prediction on the high-voltage switch operating mechanism according to the fault risk assessment value to obtain a fault prediction result.

2. The high-voltage switch operating mechanism failure prediction method according to claim 1, characterized by, The step of acquiring the current-magnetic coordination node features according to the current waveform timing features and the electromagnetic time-domain shape features comprises the following steps: acquiring a starting point time, a peak point time, and an ending point time according to the current waveform timing features; acquiring electromagnetic action stage division points according to the starting point time, the peak point time, and the ending point time, and acquiring corresponding multiple current feature values according to the electromagnetic action stage division points; acquiring a current change gradient within a preset time according to multiple current feature values; acquiring multiple touch resistances within a preset time sequence according to the current waveform timing features, acquiring a resistance influence factor according to multiple touch resistances, and acquiring a corrected current change gradient according to the resistance influence factor and the current change gradient; acquiring multiple electromagnetic force fluctuation amplitudes according to the electromagnetic time-domain shape features, acquiring multiple continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to multiple current feature values, and acquiring electromagnetic change gradients of adjacent time according to multiple continuous sampling electromagnetic force fluctuation amplitudes; dynamically associating the corrected current change gradient and the electromagnetic change gradient to obtain current-magnetic dynamic association features, and taking the current-magnetic dynamic association features as current-magnetic coordination node features.

3. The high-voltage switch operating mechanism failure prediction method according to claim 1, characterized by, The step of acquiring the vibration energy features according to the current waveform timing features and the vibration features comprises the following steps: performing time-frequency transformation on the vibration features to obtain a vibration frequency energy sequence, and performing cooperative decomposition on the vibration frequency energy sequence and the current waveform timing features by using a multivariate empirical mode decomposition algorithm to obtain multiple sets of time-frequency aligned intrinsic mode function component sets, wherein the intrinsic mode function component sets comprise vibration energy intrinsic mode components and current waveform intrinsic mode components; calculating a plurality of vibration complex correlation coefficients and a plurality of current complex correlation coefficients of the vibration energy and the current waveform on different time scales based on the vibration energy eigenmode component and the current waveform eigenmode component, and integrating the plurality of vibration complex correlation coefficients and the plurality of current complex correlation coefficients through a double-weight linear weighted fusion strategy to obtain a vibration-current fusion coefficient vector; generating a vibration energy time sequence feature through linear combination according to the vibration-current fusion coefficient vector and the vibration energy eigenmode component, and performing normalization processing on the vibration energy time sequence feature to obtain a standardized vibration energy sequence; extracting a vibration peak point and a vibration duration from the vibration feature, and obtaining a current rising edge time and a falling edge time according to the vibration peak point, the vibration duration and the current waveform time sequence feature, and obtaining a vibration event synchronization identifier according to the current rising edge time and the falling edge time; calculating a vibration energy cumulative value through a weighted energy integration algorithm based on the vibration event synchronization identifier and the standardized vibration energy sequence, and obtaining a vibration energy distribution feature according to the vibration energy cumulative value and the vibration peak point; performing feature splicing on the vibration energy distribution feature and the standardized vibration energy sequence to obtain a vibration energy feature.

4. The high-voltage switch operating mechanism failure prediction method according to claim 1, characterized by, The step of obtaining the electromagnetic driving force-motion acceleration coordination feature according to the acceleration change feature and the electromagnetic time domain morphology feature comprises: performing multi-scale decomposition on the linear displacement change feature to obtain a plurality of displacement eigenmode component sets, and obtaining a multi-scale acceleration component sequence according to the plurality of displacement eigenmode component sets; extracting an electromagnetic force gradient sequence and an electromagnetic force fluctuation frequency from the electromagnetic time domain morphology feature, and performing band self-adaptive decomposition on the electromagnetic force gradient sequence through a variational mode decomposition algorithm to obtain an electromagnetic force eigenmode component set; performing time domain alignment processing on the multi-scale acceleration component sequence and the electromagnetic force eigenmode component set based on a dynamic time warping algorithm to obtain a phase synchronization coefficient and an amplitude coupling degree between components of each scale; constructing a scale weight vector according to the phase synchronization coefficient and the amplitude coupling degree, and fusing multi-scale acceleration-electromagnetic force correlation features through an adaptive weighted fusion strategy to obtain a primary coordination feature sequence; obtaining a motion stationarity index and a displacement overshoot according to the linear displacement change feature; obtaining an electromagnetic force stable time and an overshoot amplitude according to the electromagnetic time domain morphology feature, and obtaining a motion-electromagnetic coordination degree evaluation factor according to the electromagnetic force stable time, the overshoot amplitude, the motion stationarity index and the displacement overshoot; performing dynamic correction on the primary coordination feature sequence based on the motion-electromagnetic coordination degree evaluation factor to obtain a standardized coordination feature sequence; According to the acceleration change feature, an acceleration peak time sequence and an acceleration change rate extreme point are acquired, and an acceleration-magnetic force dynamic correlation sequence is acquired according to the acceleration peak time sequence, the acceleration change rate extreme point and the electromagnetic force fluctuation frequency; the standardized collaborative feature sequence and the acceleration-magnetic force dynamic correlation sequence are tensor spliced to obtain an electromagnetic driving force-motion acceleration collaborative feature.

5. The high-voltage switch operating mechanism failure prediction method according to claim 1, characterized by, According to the step of fusing the current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature based on the lightweight Transformer model to obtain a fault risk evaluation value, comprising: The current-magnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration collaborative feature are normalized to obtain a current-magnetic coordination node feature vector, a vibration energy feature vector and an electromagnetic driving force-motion acceleration collaborative feature vector; The current-magnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector are input into a pre-trained lightweight Transformer model, wherein the lightweight Transformer model comprises an input embedding layer and a multi-layer encoder; The current-magnetic coordination node feature vector, the vibration energy feature vector and the electromagnetic driving force-motion acceleration collaborative feature vector are mapped and merged into a high-dimensional splicing feature vector based on the input embedding layer, and a corresponding vector module length is acquired according to the high-dimensional splicing feature vector, and the vector module length is taken as a fault risk evaluation value.

6. The high-voltage switch operating mechanism failure prediction method according to claim 1, characterized by, The step of predicting a fault of the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result, comprising: Acquiring historical normal operation data of the high-voltage switch operating mechanism, processing the historical normal operation data based on a lightweight Transformer model to generate a historical fault risk evaluation value sequence, and constructing a health benchmark library according to the historical fault risk evaluation value sequence, wherein the health benchmark library comprises a health benchmark interval and a health benchmark probability distribution; The fault risk evaluation value is continuously monitored to obtain a risk evaluation value sequence in a current time window; A fault decision threshold is acquired based on the health benchmark interval and the health benchmark probability distribution, and it is judged whether the risk evaluation value sequence exceeds the fault decision threshold; If the risk evaluation value sequence exceeds the fault decision threshold, it is determined that the high-voltage switch operating mechanism has a fault risk, and the fault risk is taken as a fault prediction result.

7. A high voltage switch operating mechanism failure prediction system characterized by, Comprising: The first acquisition module is configured to acquire operation data of the high-voltage switch operating mechanism in a preset time period, wherein the operation data comprises current-derived features and mechanical feedback features; The second acquisition module is configured to acquire current waveform time sequence features and electromagnetic time domain shape features according to the current-derived features, and to acquire current-magnetic coordination node features according to the current waveform time sequence features and the electromagnetic time domain shape features; a third acquisition module, configured to acquire a linear displacement change feature and a vibration feature of a moving contact in the high-voltage switch operating mechanism in an operation process according to the mechanical feedback feature, and acquire a vibration energy feature according to the current waveform time sequence feature and the vibration feature; a fourth acquisition module, configured to acquire an acceleration change feature according to the linear displacement change feature, and acquire an electromagnetic driving force-motion acceleration coordination feature according to the acceleration change feature and the electromagnetic time domain form feature; a fusion module, configured to fuse the current-electromagnetic coordination node feature, the vibration energy feature and the electromagnetic driving force-motion acceleration coordination feature based on a lightweight Transformer model to obtain a fault risk evaluation value; a prediction module, configured to perform fault prediction on the high-voltage switch operating mechanism according to the fault risk evaluation value to obtain a fault prediction result.

8. A high-voltage switch operating mechanism failure prediction system according to claim 7, characterized by The second acquisition module comprises: a first acquisition unit, configured to acquire a starting point time, a peak point time and an ending point time according to the current waveform time sequence feature; acquire electromagnetic action stage division points according to the starting point time, the peak point time and the ending point time, and acquire a plurality of current feature values corresponding to the electromagnetic action stage division points; acquire a current change gradient within a preset time according to a plurality of the current feature values; acquire a plurality of touch resistances within a preset time sequence according to the current waveform time sequence feature, acquire a resistance influence factor according to a plurality of the touch resistances, and acquire a corrected current change gradient according to the resistance influence factor and the current change gradient; acquire a plurality of electromagnetic force fluctuation amplitudes according to the electromagnetic time domain form feature, acquire a plurality of continuous sampling electromagnetic force fluctuation amplitudes within a preset time according to a plurality of the current feature values, and acquire an electromagnetic change gradient of adjacent time according to a plurality of the continuous sampling electromagnetic force fluctuation amplitudes; dynamically associate the corrected current change gradient and the electromagnetic change gradient to obtain a current-electromagnetic dynamic association feature, and take the current-electromagnetic dynamic association feature as a current-electromagnetic coordination node feature. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the method in any one of claims 1 to 6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method in any one of claims 1 to 6.

Images