A method for identifying the operating state of a pole-mounted circuit breaker based on action sequence segmentation and migration chaos features

Abstract

This invention relates to a method and system for identifying the operating state of a pole-mounted circuit breaker based on action time sequence segmentation and transfer chaotic features. It achieves high-precision time sequence segmentation by synchronously acquiring triaxial Hall signals and vibration signals, and improves adaptability to noise environments by combining transfer learning to optimize phase space reconstruction parameters. It also employs a stage-sensitive attention mechanism to fuse multi-source features, solving the technical problems of large time sequence errors, poor feature robustness, and high computational requirements of traditional methods. It has the advantages of improving segmentation accuracy, enhancing feature robustness in noisy environments to improve accuracy, and reducing computational requirements to enable real-time diagnosis of embedded devices.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring technology, specifically relating to a method for identifying the operating state of pole-mounted circuit breakers based on action timing segmentation and migration chaos features, applicable to mechanical fault diagnosis and early warning of outdoor pole-mounted circuit breakers. Background Technology

[0002] In recent years, with the continuous development of power systems and the increase in the number of power equipment, pole-mounted circuit breakers have played a crucial role in power system protection and control as key equipment. However, due to the long-term operation of pole-mounted circuit breakers and the influence of complex environmental factors, they have potential failure risks, posing a significant challenge to the safe and stable operation of power systems and potentially causing serious economic losses. Therefore, research on the operational health status of pole-mounted circuit breakers is of great significance. Summary of the Invention

[0003] The purpose of this invention is to provide a method for identifying the operating state of a pole-mounted circuit breaker based on action time sequence segmentation and migration chaotic features. This method has the advantages of improving the accuracy of time sequence segmentation, enhancing the robustness of features in noisy environments to improve accuracy, and reducing computing power requirements to achieve real-time diagnosis of embedded devices.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for identifying the operating state of a pole-mounted circuit breaker based on action sequence segmentation and migration chaos features, characterized by the following steps:

[0006] S100, synchronously acquires the three-axis Hall signal of the pole-mounted circuit breaker spindle. , , Vibration signals of the operating mechanism ;

[0007] S200, Calculate the spindle rotation angle based on the triaxial Hall signal. and angular velocity The spindle rotation angle is calculated based on the three-axis Hall signals. and angular velocity Obtain the starting point of the movement Maximum speed Maximum displacement point End point of motion Four key time points;

[0008] S300, Based on the key time points, the action is divided into four stages: the disengagement stage, the acceleration stage, the deceleration stage, and the braking stage;

[0009] S400: For vibration signals at each action stage, phase space reconstruction parameters are obtained based on transfer learning, and attractor mean distance S and Kolmogorov entropy are extracted as chaotic features.

[0010] S500 performs wavelet packet decomposition on the vibration signals at each stage of the motion and calculates the energy entropy. E As a temporal energy characteristic;

[0011] S600 integrates chaotic features and temporal energy features from four action stages, and generates a weighted feature vector through a stage-sensitive attention mechanism. ;

[0012] S700: Input the weighted feature vector into a lightweight support vector machine classifier and output the operation state category and fault location result;

[0013] S800: Calculate the anomaly level based on the characteristic deviation β and trigger an early warning.

[0014] The present invention further provides that the spindle rotation angle The calculation formula is: in , , The output signals are from three Hall sensors spatially distributed 120° apart; the angular velocity is calculated as follows: The present invention further specifies that the method for determining the key time point is as follows:

[0015] Starting point of movement ;

[0016] Maximum speed ;

[0017] Maximum displacement point ;

[0018] End point of motion ;

[0019] in () represents the maximum angular velocity throughout the entire journey; The maximum rotation angle throughout the entire stroke; based on the above time points, the motion state is divided into the disengagement phase, acceleration phase, deceleration phase, and braking phase, that is, the disengagement phase time period is [0, The acceleration phase is [ ], The deceleration phase is [ ], The braking phase is [ ].

[0020] The present invention further specifies that the phase space reconstruction step is as follows: acquiring the vibration signal time series of four stages. Their lengths are denoted as n, k = 1, 2, 3, 4…, and the sampling interval is… Calculate the mutual information value given different delay times τ, and plot the mutual information curve. Then, determine the value based on the first minimum point of the curve. To select the optimal delay time, the embedding dimension is calculated using the spurious neighbor method, and the minimum embedding dimension is selected as the optimal value. Construct a delay vector , which is: ,in , , Let M be the phase point in phase space, and M be the number of phase points.

[0021] The present invention further includes a transfer learning process for the phase space reconstruction parameters during the braking phase, which is achieved through the following formula: , The values ​​of a and b are 0.2 and 0.1, respectively. The standard deviation of outdoor vibration signal noise is calculated as follows: The time interval from the stored vibration signal is […]. All data points between ] Then calculate the mean and the standard deviation based on the mean. .

[0022] The invention further involves reconstructing the phase space of the vibration time series at each stage based on the delay time and embedding dimension to construct a 3D phase space chaotic attractor morphology diagram; then, calculating the distance matrix between each attractor and their average distance S; finally, selecting an arbitrary reference attractor from the reconstructed phase space. ,calculate Other attractors The Euclidean distance is used to sum the weighted contributions of all attractors to obtain the correlation integral C(m,r); a suitable critical distance r is chosen to calculate the correlation dimension D(m) in the scale-free interval.

[0023] The present invention further includes calculating the K-entropy using the correlation integral method, continuously decreasing the value of r during the calculation process, and calculating the value of C(m,r). When C(m,r) does not change... Other attractors Determine the K-entropy when it changes with the Euclidean distance.

[0024] The present invention further includes performing four-level wavelet packet decomposition on the vibration signals at each stage to obtain sub-frequency band signals, and calculating the energy of each sub-frequency band signal. Then, the energy percentage of each sub-band signal is calculated. The energy entropy E of each stage is calculated based on the energy percentage, where: , is the discrete sampled signal of the nth sub-band; f is the index of the discrete sample point; The sampling interval; ; ;

[0025] The present invention further includes obtaining features at each stage. Attention weights are calculated to obtain corresponding weights, and weighted feature vectors are calculated based on these weights. The present invention further includes a calculation method for the average attractor distance S across the four stages when calculating the characteristic deviation β, using the following formula: Then determine the overall anomaly level and select... The maximum value in the value is taken as the overall anomaly degree as β, where: when β < 2.0, the warning level is normal; when 2.0 ≤ β < 3.0, the warning level is attention; when 3.0 ≤ β < 4.0, the warning level is warning; when β ≥ 4.0, the warning level is crisis.

[0026] The beneficial effects of this invention are as follows: This application provides a method and system for identifying the operating state of a pole-mounted circuit breaker based on action time sequence segmentation and transfer chaotic features. It achieves high-precision time sequence segmentation by synchronously acquiring triaxial Hall signals and vibration signals, optimizes phase space reconstruction parameters by combining transfer learning to improve adaptability to noise environments, and uses a stage-sensitive attention mechanism to fuse multi-source features. This solves the technical problems of large time sequence errors, poor feature robustness, and high computational requirements of traditional methods. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0028] Figure 1 This is a schematic diagram of the process of an embodiment of the present invention. Detailed Implementation

[0029] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0030] In existing technologies, monitoring the operating status of pole-mounted circuit breakers often employs high-speed cameras or contact sensors. However, these methods are susceptible to the effects of light, temperature, and electromagnetic interference in complex outdoor environments, leading to insufficient timing segmentation accuracy. For example, traditional high-speed camera methods may produce timing errors as high as ±0.5 milliseconds under strong light or shadow conditions, while the installation of contact sensors may alter the mechanical dynamics of the circuit breaker. Furthermore, existing feature extraction methods exhibit poor robustness in noisy environments; for instance, the accuracy of fixed phase space reconstruction parameters drops significantly under low signal-to-noise ratio conditions, and the energy transfer characteristics during the operating phase are not considered, resulting in ambiguous fault location. Simultaneously, the computational complexity of high-dimensional chaotic features limits their application in embedded devices.

[0031] To address the aforementioned issues, a non-contact time-series segmentation method needs to be designed for outdoor operating conditions to improve the noise adaptability of feature extraction and reduce computational complexity. By analyzing the physical characteristics during circuit breaker operation, it was found that the coupling relationship between the spindle rotation angle and the vibration signal can reflect the differences in energy distribution at different stages. Furthermore, considering the parameter transfer requirements between laboratory models and field data, a noise-adaptive phase space reconstruction method is proposed. Simultaneously, by fusing multi-stage features and introducing a lightweight classification mechanism, the compatibility issue between high-dimensional features and low-computing-power platforms is resolved.

[0032] Therefore, this application proposes the following technical solution: synchronously acquire the three-axis Hall signal of the spindle and the vibration signal of the operating mechanism, determine four key time points by calculating the rotation angle and angular velocity, and divide the disengagement, acceleration, deceleration and braking stages; extract chaotic features and time-series energy features from the vibration signals of each stage, and optimize the phase space reconstruction parameters by combining transfer learning; fuse features through a stage-sensitive attention mechanism, use a lightweight classifier to realize state identification and fault location, and trigger graded early warning based on feature deviation.

[0033] The specific technical solution is as follows:

[0034] See Figure 1 This invention provides a method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features, comprising the following steps:

[0035] S100, synchronously acquires the three-axis Hall signal of the pole-mounted circuit breaker spindle. , , Vibration signals of the operating mechanism ;

[0036] S200, Calculate the spindle rotation angle based on the triaxial Hall signal. and angular velocity The spindle rotation angle is calculated based on the three-axis Hall signals. and angular velocity Obtain the starting point of the movement Maximum speed Maximum displacement point End point of motion Four key time points;

[0037] S300, Based on the key time points, the action is divided into four stages: the disengagement stage, the acceleration stage, the deceleration stage, and the braking stage;

[0038] S400: For vibration signals at each action stage, phase space reconstruction parameters are obtained based on transfer learning, and attractor mean distance S and Kolmogorov entropy are extracted as chaotic features.

[0039] S500 performs wavelet packet decomposition on the vibration signals at each stage of the motion and calculates the energy entropy. E As a temporal energy characteristic;

[0040] S600 integrates chaotic features and temporal energy features from four action stages, and generates a weighted feature vector through a stage-sensitive attention mechanism. ;

[0041] S700: Input the weighted feature vector into a lightweight support vector machine classifier and output the operation state category and fault location result;

[0042] S800 calculates the anomaly level and triggers an early warning based on the characteristic deviation β. The triaxial Hall signal refers to three non-contact magnetic induction signals spatially distributed 120° apart, specifically implemented using an orthogonally arranged Hall sensor array. Its output signal, after geometric synthesis, accurately reflects the spatial angle changes of the main axis. Key time point division is based on dynamic threshold detection of angular velocity and angle. Notably, a current sensor monitors the coil current; when the current sensor detects current, it is recorded as the zero-point of acquisition. For the triaxial Hall signal acquisition system:

[0043] Sensor Selection and Installation: Three linear Hall effect sensors (recommended model: Allegro MicroSystems A1324LUA-T) are used, spatially positioned 120° apart orthogonally mounted on a non-magnetic retaining ring. This retaining ring is fitted onto the end of the circuit breaker spindle, ensuring synchronous rotation with the spindle without affecting its mechanical movement. A constant air gap of 1-2mm is maintained between the sensors and the spindle magnet (permanent magnet) to ensure the magnetic field strength is within the sensor's linear operating range.

[0044] Signal principle and conditioning: Each Hall sensor outputs a voltage signal proportional to the magnetic field strength at its location. The signal is amplified by an instrumentation amplifier (such as AD620), processed by an anti-aliasing low-pass filter (cutoff frequency set to 1kHz), and then sent to the data acquisition card.

[0045] Purpose of data acquisition: To accurately and non-contactly calculate the spindle's rotation angle and angular velocity, providing a high-precision time-domain reference for subsequent time-series segmentation.

[0046] For vibration signal acquisition systems

[0047] Sensor Selection and Installation: An industrial-grade ICP piezoelectric accelerometer (recommended model: PCBPiezotronics 608A11) is used, with a frequency response range (0.5Hz - 10kHz) covering the characteristic frequencies of shocks and vibrations generated by the circuit breaker operating mechanism. The sensor is securely mounted on the rigid surface of the operating mechanism housing using a steel magnetic base, and the mounting position should be as close as possible to moving parts (such as the spindle bearing housing or mechanism linkage) to ensure effective coupling of vibration energy.

[0048] Signal Principle and Conditioning: The sensor outputs a charge signal proportional to the vibration acceleration, which is converted into a low-impedance voltage signal via a built-in ICP circuit. This signal is then passed through an anti-aliasing low-pass filter (cutoff frequency set to 5kHz) before being connected to the data acquisition card.

[0049] Purpose of data collection: To extract chaotic features and temporal energy features related to the mechanical state at each action stage, which is the core information source for fault diagnosis.

[0050] Coil current signal acquisition system

[0051] Sensor selection and installation: Use a closed-loop Hall current sensor (recommended model: LEM LAH 50-P), with its primary wire connected in series in the closing / opening coil circuit, and the sensor body fixedly installed.

[0052] Signal principle and conditioning: The sensor outputs a voltage signal that is proportional to the current of the measured coil. This signal is then conditioned and connected to the data acquisition card.

[0053] Purpose of data acquisition: To accurately determine the absolute time starting point (t=0) of the operation process, i.e., the moment the coil is energized. This method can supplement or verify the angular velocity threshold method for determining the start point of motion, further enhancing the reliability of the timing starting point.

[0054] 4. Synchronous data acquisition scheme

[0055] All sensor signals are sampled simultaneously by the same multi-channel synchronous data acquisition card (such as National Instruments NI-9220), with the sampling rate uniformly set to 10 kHz to ensure strict synchronization between all signal channels and avoid time delay errors. Acquisition is controlled and triggered by host computer software (such as LabVIEW), and the trigger signal can be the rising edge of the coil current.

[0056] Furthermore, this application defines the starting point of motion as the first time the angular velocity exceeds the maximum value by 1%, which eliminates false triggering caused by noise. The action phase division is based on energy transfer characteristics, decomposing the continuous motion process into four physically meaningful intervals for targeted feature extraction. Transfer learning optimization of phase space reconstruction parameters refers to dynamically adjusting the delay time and embedding dimension based on the noise statistical characteristics of the braking phase. For example, the laboratory calibration parameters are corrected using the noise standard deviation to improve the attractor reconstruction accuracy in outdoor environments. Chaotic feature extraction includes the attractor average distance and Kolmogorov entropy. The former reflects the degree of clustering of phase space trajectories, while the latter characterizes the dynamic complexity of the system; combining the two enhances the ability to represent state anomalies. Temporal energy features are obtained through wavelet packet decomposition, converting the vibration signal energy distribution into an entropy index to quantify energy transfer anomalies in different frequency bands. The phase-sensitive attention mechanism dynamically allocates weights based on the contribution of each stage's features to the fault; for example, it automatically focuses on the abnormal stage through the difference in entropy values ​​after normalization.

[0057] Specifically, this method first acquires the spindle motion trajectory through non-contact measurement of triaxial Hall signals, avoiding interference from traditional contact sensors on the mechanical system. Based on dynamic threshold detection of angular velocity and angle, it accurately divides the action into four stages, for example, using velocity extreme points and displacement extreme points to define the boundaries between acceleration and deceleration stages. Addressing outdoor vibration and noise issues, background noise statistical characteristics are extracted during the braking stage, and the phase space parameters calibrated in the laboratory are linearly corrected, for example, by increasing the delay time by 0.2 times the noise standard deviation. For vibration signals at each stage, the phase space is reconstructed, and the average attractor distance and Kolmogorov entropy are calculated. Simultaneously, a four-level wavelet packet decomposition is performed to obtain 16 sub-band energy entropies. An attention mechanism is used to calculate the weight coefficients of features at each stage, for example, assigning higher weights to stages with abnormally high entropy values, ultimately fusing them into a three-dimensional weighted feature vector. This vector is input into a lightweight support vector machine, which outputs the state category and fault location results, and triggers graded early warnings based on the feature deviation compared to a preset threshold.

[0058] Compared to existing technologies, traditional methods suffer from significant temporal segmentation errors due to noise interference in outdoor environments. This solution, however, improves temporal accuracy to the microsecond level through non-contact Hall signals and dynamic threshold detection. Existing fixed-parameter phase space reconstruction is prone to generating spurious attractors under low signal-to-noise ratio conditions. This solution improves feature extraction accuracy by approximately 15% at the same noise level through noise-adaptive parameter migration. Furthermore, traditional methods require computation of chaotic features with more than 12 dimensions, while this solution compresses the feature dimension to 3 dimensions through a staged attention mechanism, reducing computation time by approximately 60%, making it more suitable for embedded devices.

[0059] Through the above technical solutions, this application effectively solves the three core problems of condition monitoring of pole-mounted circuit breakers in complex outdoor environments: non-contact Hall signals and dynamic threshold detection improve the accuracy and reliability of time-series segmentation; noise-adaptive phase space parameter migration enhances the robustness of chaotic features; and multi-stage feature fusion and lightweight classification mechanisms significantly reduce computational resource requirements while ensuring diagnostic accuracy. This method provides a feasible technical path for online monitoring of outdoor circuit breakers, and is particularly suitable for real-time condition assessment and early warning in power grid distribution automation scenarios.

[0060] This application further specifies the spindle rotation angle. The calculation formula is: in , , The output signals are from three Hall sensors spatially distributed 120° apart; the angular velocity is calculated as follows: .

[0061] Among them, the Hall sensor with a spatial distribution difference of 120° refers to three sensors that are evenly distributed in space at 120° intervals. Specifically, this can be achieved by installing them in a ring array, and the spatial symmetrical layout eliminates measurement errors in a single direction.

[0062] The arctangent function in the rotation angle calculation formula is used to convert the triaxial magnetic field signal into a plane angle. Specifically, it can be implemented using a four-quadrant arctangent algorithm. The angle calculation from 0 to 360° without dead angles is achieved by judging the signs of the numerator and denominator.

[0063] Among them, the differential operation in the angular velocity calculation is used to capture the dynamic characteristics of the main shaft motion. Specifically, it can be implemented using the five-point center difference method, which calculates the instantaneous angular velocity through the rate of change of angle between adjacent sampling points.

[0064] Specifically, the spatial distribution differences of the output signals from three Hall sensors are used to construct a mapping relationship between magnetic field strength and mechanical angle. During signal processing, the three-axis magnetic field components are substituted into trigonometric function formulas for vector synthesis, eliminating the influence of environmental magnetic field interference on a single sensor. Angular velocity is obtained through time differentiation of the angle signal, accurately reflecting the acceleration change process of the main shaft motion. The resulting angle and angular velocity data are used for subsequent key time point determination, providing a quantitative basis for the division of action stages.

[0065] Compared to existing technologies, traditional high-speed camera methods rely on optical measurements and are susceptible to interference from outdoor lighting, while contact-based travel sensors can alter the mechanical characteristics of circuit breakers. This solution employs non-contact magnetic field measurement, achieving full-circuit angle resolution through a triaxial Hall sensor array, thus avoiding the influence of mechanical installation on moving parts.

[0066] Through the above technical solution, this application effectively solves the problem of accurately measuring the spindle motion parameters in complex outdoor electromagnetic environments, providing high-precision time-domain characteristic data for subsequent action phase division. The non-contact measurement method avoids installation interference from traditional mechanical sensors, while utilizing the spatial complementarity of triaxial signals enhances anti-interference capabilities, establishing a reliable time-series segmentation basis for circuit breaker operating state identification.

[0067] This application further proposes the following method for determining key time points:

[0068] Starting point of movement ;

[0069] Maximum speed ;

[0070] Maximum displacement point ;

[0071] End point of motion ;

[0072] in () represents the maximum angular velocity throughout the entire journey; The maximum rotation angle throughout the entire stroke; based on the above time points, the motion state is divided into the disengagement phase, acceleration phase, deceleration phase, and braking phase, that is, the disengagement phase time period is [0, The acceleration phase is [ ], The deceleration phase is [ The braking phase is [ ], where 0 is the time point when the coil is energized. In this application, the energization state of the coil is detected by a current sensor.

[0073] The closing process is divided into four stages.

[0074] 1) Tripping stage. The current in the closing coil appears, the electromagnet core strikes and trips, the closing spring holding stop is released, and the shaft pin is released, thereby generating a vibration signal. At this time, the moving contact has not started to move.

[0075] 2) Acceleration phase. The closing energy storage spring is released from its constraint, releasing energy. The main shaft crank arm rotates clockwise, and the vibration signal is generated by the contact action of the mechanism components. The moving contact speed increases to the maximum, corresponding to...

[0076] 3) Deceleration phase. The vibration signal encompasses the violent collision between the moving and stationary contacts, the compression stroke, the reduction of the moving contact speed to zero, and the compression of the trip spring to store energy.

[0077] 4) Braking stage. This stage is the contact bounce process of the closing mechanism. The contact bounce time is an effective parameter for measuring the mechanical characteristics of the circuit breaker. When the main shaft crank arm rotates to the end of its stroke, the closing holding stop locks the shaft pin. The vibration signal is reflected as the energy attenuation change of the operating mechanism from moving to stationary. Although the main shaft crank arm stops moving, the vibration signal has not yet ended.

[0078] By identifying the different stages in the circuit breaker closing process and analyzing the vibration signals of each stage, the local characteristics of each sub-stage can be captured, allowing for a better understanding of the local behavior and characteristics of the signals. This further helps to pinpoint the sub-stage in which the fault occurs, thereby locating the specific location of the fault and improving the accuracy of fault diagnosis.

[0079] This application further proposes the following steps for phase space reconstruction: acquiring vibration signal time series for four stages. Their lengths are denoted as n, k = 1, 2, 3, 4…, and the sampling interval is… Calculate the mutual information value given different delay times τ, and plot the mutual information curve. Then, determine the value based on the first minimum point of the curve. To select the optimal delay time, the embedding dimension is calculated using the spurious neighbor method, and the minimum embedding dimension is selected as the optimal value. Construct a delay vector , which is: ,in , , Let M be the number of phase points in phase space; the optimal delay time is calculated using the mutual information method based on Shannon's information entropy theory. Mutual information is defined as... I ( τ ): Wherein: H is the average information content of the system for the variables; In the original time series The probability of occurrence; After a delay of time τ, it is denoted as ; This represents the probability of its occurrence in the delayed time series. for and The probability of simultaneous occurrence. The τ corresponding to the first minimum value of the mutual information curve I(τ) is the optimal delay time.

[0080] The spurious neighbor method used in this application is as follows: Record phase points The nearest neighbor is The Minkowski distance between the two phase points is:

[0081] In the formula: q The value is 2, and the embedding dimension is continuously increased. m The value is calculated, and the corresponding number of false nearest neighbors is counted until the number of false nearest neighbors approaches 0 (proportion < 5%), or the number remains unchanged. At this point, the minimum value is... m This is the optimal embedding dimension. Noise standard deviation. This refers to the intensity index of random interference in the vibration signal during the braking phase. Specifically, it can be obtained by calculating the degree to which data points deviate from the mean. This index can quantify the impact of the outdoor environment on the signal, especially during the braking phase, where its outdoor influence is significant. Therefore, transfer learning is performed on the phase space reconstruction parameters for the braking phase, which is achieved through the following formula: , The values ​​of a and b are 0.2 and 0.1, respectively. The standard deviation of outdoor vibration signal noise is calculated as follows: The time interval from the stored vibration signal is […]. All data points between ] Then calculate the mean and the standard deviation based on the mean. .

[0082] Specifically, in the vibration signal analysis process during the braking phase, the signal segment corresponding to that phase is first extracted from historical data, such as the time interval […]. All sampling points within the specified range. By calculating the mean of this data segment, the average of the squared deviations of each data point from the mean is further calculated, and then the square root is taken to obtain the standard deviation. The baseline delay time was determined based on the laboratory environment. and embedding dimension Each is superimposed with a correction term related to noise intensity. and The coefficients a and b can be 0.2 and 0.1, respectively. This yields the migration parameters for adapting to outdoor noise environments. and This is used for subsequent phase space reconstruction. This process effectively solves the problem of parameter mismatch between laboratory models and field applications.

[0083] This application further proposes to reconstruct the phase space of the vibration time series at each stage based on the delay time and embedding dimension, constructing a 3D phase space chaotic attractor morphology diagram; then, the distance matrix between each attractor is calculated, and then their average distance S is calculated; finally, an arbitrary reference attractor is selected from the reconstructed phase space. ,calculate Other attractors The Euclidean distance is used to sum the weighted contributions of all attractors to obtain the correlation integral C(m,r); a suitable critical distance r is chosen to calculate the correlation dimension D(m) in the scale-free interval.

[0084] Wherein: The formula for calculating the average distance S is: ;

[0085] calculate Other attractors The Euclidean distance is expressed by the following formula: Related integrals: In the above formula, r is the radius (critical distance) of the m-dimensional hypersphere, and ∂ is the heaviside step function: Select an appropriate critical distance r The correlation dimension exists in the scale-free interval. D ( m ): ;

[0086] Subsequently, K-entropy is calculated using the correlation integral method. During the calculation, the value of r is continuously decreased, and the value of C(m,r) is calculated. When C(m,r) does not change... Other attractors When the Euclidean distance changes, determine the K-entropy. .

[0087] This application further proposes to perform four-level wavelet packet decomposition on the vibration signals at each stage to obtain sub-frequency band signals, and to calculate the energy of each sub-frequency band signal. Then, the energy percentage of each sub-band signal is calculated. The energy entropy E of each stage is calculated based on the energy percentage, where: , is the discrete sampled signal of the nth sub-band; f is the index of the discrete sample point; The sampling interval; ; ;

[0088] This application further proposes a fusion feature for dynamic features, namely, for the chaotic features S, K and energy entropy E of each of the four stages, that is, to obtain the features of each stage. Then construct the feature vector: f=[ ], 1 represents the tripping stage, 2 represents the acceleration stage, 3 represents the deceleration stage, and 4 represents the braking stage;

[0089] Then calculate the stage-sensitive attention weights: ;

[0090] In the above formula =0.5, Let be the 3-dimensional feature vector of the i-th stage. This is a reference characteristic for the normal state;

[0091] Calculate the weighted eigenvector based on the weights. .

[0092] Then the weighted feature vectors Input a lightweight support vector machine classifier and output the operation state category and fault location result; specifically as follows: classify using a lightweight SVM, using a pre-trained sparse SVM model (feature selection set S is determined by L1 regularization). , The weights are the weights of the L1-regularized support vector machine, and its decision function is: in This is the RBF kernel function.

[0093] The training used 500 sets of samples, including historical normal operation data and known fault data. The feature vector is a weighted 3-dimensional feature.

[0094] L1 regularized linear SVM is used for feature selection. The sparsity parameter is determined by 5-fold cross-validation. Finally, features with weight coefficients greater than 0.05 and the largest weight are retained.

[0095] The kernel function used is the RBF kernel, and the hyperparameters and penalty coefficients are optimized through grid search.

[0096] After training, the model supports deployment on embedded devices, with an inference time of less than 5ms.

[0097] Fault stage location is as follows: Spring fatigue: If S < μ during acceleration stage S −2σ S (μ and σ are normal sample statistics); Core jamming: If K > μ during the braking phase K +2σ K Axle pin wear: If the energy entropy E > μ during the deceleration phase E +3σ EThe system can also identify the following fault types: Tripping stage: brake wear (characteristic: abnormally high energy entropy); Acceleration stage: spring breakage (characteristic: abnormally low attractor distance); Braking stage: buffer failure (characteristic: significant increase in K entropy); If identified as spring fatigue, the system will further locate the specific spring group (such as opening spring / closing spring) through energy distribution analysis during the acceleration stage.

[0098] The average attractor distance S in the four stages is calculated when calculating the characteristic deviation β, and the formula is as follows: Then determine the overall anomaly level and select... The maximum value in the range is taken as the overall anomaly score, β, where

[0099] When β < 2.0, the warning level is normal; when 2.0 ≤ β < 3.0, the warning level is alert; when 3.0 ≤ β < 4.0, the warning level is warning; when β ≥ 4.0, the warning level is crisis.

[0100] This application addresses the pain points of poor outdoor environmental adaptability and high noise interference of pole-mounted circuit breakers by using multi-sensor low-coupling temporal alignment and chaotic feature transfer learning, achieving both high precision and low deployment cost.

[0101] As used in the specification and claims, certain terms refer to specific components. Those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The term "comprising" throughout the specification and claims is an open-ended term and should be interpreted as "comprising but not limited to." "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error.

[0102] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for identifying the operating state of a pole-mounted circuit breaker based on action sequence segmentation and migration chaos features, characterized in that, Includes the following steps: S100, synchronously acquires the three-axis Hall signal of the pole-mounted circuit breaker spindle. , , Vibration signals of the operating mechanism ; S200, Calculate the spindle rotation angle based on the triaxial Hall signal. and angular velocity The spindle rotation angle is calculated based on the three-axis Hall signals. and angular velocity Obtain the starting point of the movement Maximum speed Maximum displacement point End point of motion Four key time points; S300, Based on the key time points, the action is divided into four stages: the disengagement stage, the acceleration stage, the deceleration stage, and the braking stage; S400. For the vibration signals at each action stage, phase space reconstruction parameters are obtained based on transfer learning. The average attractor distance S and Kolmogorov entropy are extracted as chaotic features. The transfer learning of phase space reconstruction parameters for the braking stage is achieved through the following formula: , The values ​​of a and b are 0.2 and 0.1, respectively. The standard deviation of outdoor vibration signal noise is calculated as follows: The time interval from the stored vibration signal is […]. All data points between ] Then calculate the mean and the standard deviation based on the mean. ; S500 performs wavelet packet decomposition on the vibration signals at each stage of the motion and calculates the energy entropy. E As a temporal energy characteristic; S600 integrates chaotic features and temporal energy features from four action stages, and generates a weighted feature vector through a stage-sensitive attention mechanism. That is, to obtain the features of each stage. Attention weights are calculated to obtain corresponding weights, and weighted feature vectors are calculated based on these weights. ; S700: Input the weighted feature vector into a lightweight support vector machine classifier and output the operation state category and fault location result; S800. Calculate the anomaly level and trigger an early warning based on the characteristic deviation β. When calculating the characteristic deviation β, the average attractor distance S for the four stages is calculated using the following formula: ; Then determine the overall anomaly level and select... The maximum value in the range is taken as the overall anomaly score, β, where When β < 2.0, the warning level is normal; When 2.0 ≤ β < 3.0, the warning level is "Caution"; When 3.0 ≤ β < 4.0, the warning level is a warning; when β ≥ 4.0, the warning level is a crisis.

2. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 1, characterized in that, The spindle rotation angle The calculation formula is: in , , The output signals are from three Hall sensors spatially distributed 120° apart; the angular velocity is calculated as follows: .

3. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 1, characterized in that, The method for determining key time points is as follows: Start point of movement Maximum speed ; Maximum displacement point ; End point of motion ; in () represents the maximum angular velocity throughout the entire journey; The maximum rotation angle throughout the entire stroke; based on the above time points, the motion state is divided into the disengagement phase, acceleration phase, deceleration phase, and braking phase, that is, the disengagement phase time period is [0, The acceleration phase is [ ], The deceleration phase is [ ], The braking phase is [ ].

4. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 1, characterized in that, The phase space reconstruction steps are as follows: Obtain the vibration signal time series for four stages. Their lengths are denoted as n, k = 1, 2, 3, 4…, and the sampling interval is… Calculate the mutual information value given different delay times τ, and plot the mutual information curve. Then, determine the value based on the first minimum point of the curve. The optimal delay time was selected. The embedding dimension is calculated using the spurious neighbor method, and the minimum embedding dimension is selected as the value. Construct a delay vector , which is: ,in , , Let M be the phase point in phase space, and M be the number of phase points.

5. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 1, characterized in that, Phase space reconstruction is performed on the vibration time series of each stage based on the delay time and embedding dimension to construct a 3D phase space chaotic attractor morphology diagram; then the distance matrix between each attractor is calculated, and their average distance S is calculated; finally, an arbitrary reference attractor is selected from the reconstructed phase space. ,calculate Other attractors The Euclidean distance is used to sum the weighted contributions of all attractors to obtain the correlation integral C(m,r); a suitable critical distance r is chosen to calculate the correlation dimension D(m) in the scale-free interval.

6. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 5, characterized in that, The Kolmogorov entropy is calculated using the correlation integral method. During the calculation, the value of r is continuously decreased, and the value of C(m,r) is calculated. When C(m,r) does not change... Other attractors When the Euclidean distance changes, determine the Kolmogorov entropy, where r is the radius of the m-dimensional hypersphere.

7. The method for identifying the operating state of a pole-mounted circuit breaker based on action timing segmentation and migration chaotic features according to claim 1, characterized in that, The vibration signals at each stage are decomposed into sub-band signals using a four-level wavelet packet decomposition method, and the energy of each sub-band signal is calculated. Then, the energy percentage of each sub-band signal is calculated. The energy entropy E of each stage is calculated based on the energy percentage, where: ， is the discrete sampled signal of the Kth sub-band; f is the index of the discrete sample point; The sampling interval; ; 。

Images