Device and method for detecting mechanical state of high-voltage isolating switch

Abstract

The invention relates to a high-voltage isolation switch mechanical state detection device and detection method, and belongs to the technical field of power equipment detection. The system mainly comprises a power source monitoring module, a reference state mapping module, a first analysis module, a second analysis module, a fault interpretation module, a state evaluation module, a verification module and a calibration module. Through the steps of dynamic characteristic model construction, multi-dimensional signal processing, geometric analysis, fault interpretation and the like, precise monitoring and diagnosis of internal deformation, overall displacement and absolute attitude change of the disconnecting switch are realized. The method overcomes the defects of a traditional method in the aspects of application range, complexity and reliability, has the advantages of being high in anti-interference capability, high in measurement precision, high in real-time performance and the like, and is suitable for full-life-cycle management of the high-voltage switchgear.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment detection, and particularly relates to a high-voltage disconnecting switch mechanical state detection device and a detection method. BACKGROUND

[0002] The high-voltage disconnecting switch is a key device of the power system, most of which works in an outdoor environment. Due to uneven manufacturing processes, variable operating environments, untimely or inadequate maintenance, and the like, various faults are prone to occur. Among them, mechanical fault defects account for more than 50%, which seriously affects the safe and stable operation of the power grid. As the most widely used high-voltage switching device in the power system, the high-voltage disconnecting switch is mainly applied to each substation and switching station, and is installed on both sides of the circuit breaker, used to complete the switching operation with the circuit breaker, change the operation mode, isolate the power supply, or connect and disconnect the small load device. When the operation mode is switched or the equipment is powered off for maintenance, the disconnecting switch has a clear visible break after being opened, which can ensure the safe isolation of the powered-off equipment and line from the operating system, thereby ensuring the safety of the maintenance personnel and the equipment.

[0003] The traditional mechanical state monitoring method of the high-voltage disconnecting switch includes image recognition, optical sensing, pressure monitoring, monitoring method based on auxiliary contact principle, and monitoring method based on attitude sensing technology and micro switch technology. However, these methods have certain limitations in practical application: the image recognition method has a high degree of automation and intuitive judgment process, but the recognition accuracy is low, is greatly affected by weather, has high cost, and is easily affected by magnetic field and environmental interference; the optical sensing method is simple to measure and has strong anti-strong electromagnetic field interference ability, but has high installation precision requirement, is difficult to adjust on site, is affected by factors such as light transmission distance, rain, snow, humidity, and visibility, and has low reliability and accuracy; the pressure monitoring method is simple, reliable, and highly practical, but has limited application range and is only suitable for disconnecting switches with handling springs; the monitoring method based on the auxiliary contact principle has a simple principle, but there are problems of signal transmission and auxiliary switch function in the high-voltage area, and abnormal transmission links in the low-voltage area may affect the judgment accuracy; the monitoring method based on attitude sensing technology uses microelectromechanical technology, and the sensor has low cost, light weight, low power consumption, high stability, and is easy to integrate, but the early attitude sensor has high cost, large size, complex structure, and poor stability; the micro switch technology has small size, high precision, and reliable data, but may be affected by the collision of mechanical moving parts.

[0004] Therefore, it is of important practical significance and application value to develop a high-voltage disconnecting switch mechanical state detection technology that can overcome the above limitations. SUMMARY

[0005] To overcome the problems in the background art, the present invention provides a high-voltage disconnector mechanical condition detection device and detection method. Through the coordinated operation of the above modules, the present invention achieves comprehensive monitoring and accurate diagnosis of the mechanical condition of the high-voltage disconnector, solving the shortcomings of traditional methods in terms of applicable scope, installation and commissioning complexity and reliability. At the same time, it has the characteristics of strong anti-interference ability, high measurement accuracy and high real-time performance, and is suitable for the full life cycle management of various high-voltage switchgear.

[0006] To achieve the above objectives, the present invention provides a technical solution as follows: a high-voltage disconnector mechanical condition detection device, characterized in that it comprises: The power source monitoring module is used to construct a dynamic characteristic model of the motor's operating state based on real-time collected drive motor voltage and current data. This model includes time series nodes representing the instantaneous power change of the motor, electrical parameter edges connecting nodes within the same operating cycle, and time sequence edges connecting nodes in different operating cycles. Instantaneous power is extracted from the original voltage and current data, and time series nodes are generated by using this instantaneous power data through threshold triggering or time window averaging. The weights of the electrical parameter edges are set according to structural adjacency and electrical coupling, and the time sequence edges are based on time series analysis or Markov chain models. Calculation of electrical parameter edges Geometric topological mapping: Mathematically, the weight of an edge can be quantified by equivalent mechanical impedance or energy transfer efficiency, for example: Where Zmech is the mechanical impedance and θij is the transmission angle between nodes. The calculation results need to be coupled with electrical parameters (such as current phase) to reflect the "electromechanical" coupling characteristics.

[0007] Numerical calculation example: Taking gear transmission as an example, the electrical parameter side can be represented as: Where ω is the angular velocity, Irms is the effective value of the current, and the coefficient k is determined by the gear ratio and motor parameters.

[0008] Markov chain training and implementation Training Algorithm: The training process for the Markov chain model used in the temporal edges includes: 1. State space definition: Discretize time series nodes into states (such as "high power", "low power", "abnormal fluctuation").

[0009] 2. Transition Probability Estimation: The state transition matrix P is calculated using maximum likelihood estimation (MLE), where the elements... Nij represents the number of transitions from state i to state j.

[0010] 3. Convergence optimization: Use iterative algorithms (such as Power Iteration) to solve for the stationary distribution to ensure model convergence.

[0011] Parameter optimization: Hyperparameters such as the number of states and transition order need to be adjusted through Bayesian optimization or grid search. The loss function is often chosen as the negative log-likelihood (NLL) or the BIC criterion.

[0012] Data requirements: Training requires a large amount of historical operating cycle data (recommended >10^4 sequences), and the data must cover normal and typical failure modes (such as jamming, wear).

[0013] Model integration and dynamic updates Deep learning integration: Dynamic feature models are often combined with deep learning, for example, using LSTM to encode time series nodes, using GNN to process electrical parameter edge topology, and outputting state classification probabilities.

[0014] Online learning mechanism: The model needs to support online updates, and optimize the transition probability through retraining on false positive samples (such as incremental learning of SVM) or reinforcement learning (reward function is the fault detection accuracy).

[0015] The baseline state mapping module 2 maps the nodes in the dynamic characteristic model 9 to the standard health state reference network 10. For example... Figure 3 As shown, the standard health status reference network 10 consists of multiple reference points with predetermined relative geometric positions, and each monitoring point node is mapped to three reference points. For example, node A is mapped to reference points P1, P2, and P3, where the geometric relationships between P1, P2, and P3 are known and fixed. During the mapping process, the baseline state mapping module 2 determines the mapping relationship by calculating the Euclidean distance or angular deviation between the node and the reference points and outputs it to the second analysis module 4. The second analysis module 4 calculates the temporal variation index of the spatiotemporal relative geometric relationship based on the mapping relationship, such as the rate of change or cumulative change of the relative distance or relative angle defined by the monitoring point node and its mapped set of reference points. These indices are used for subsequent fault interpretation.

[0016] The baseline state mapping module is used to map nodes in the dynamic characteristic model to reference points of a predefined standard health state reference network, establishing a mapping relationship. The standard health state reference network is established based on a large amount of historical health data, expert experience, or simulation models. The first analysis module is configured to use multi-dimensional signal processing algorithms such as graph neural networks or time series analysis to analyze the dynamic characteristic model, outputting a weight matrix representing the correlation of electrical parameters between nodes and feature vectors encoding node states and neighborhood information. The second analysis module is configured to process the mapping relationship and output a time change index representing the spatiotemporal relative geometric relationship of the monitoring point with respect to the reference point of the standard health state reference network through geometric analysis. The fault interpretation module is used to perform the following steps: Step S1, based on the time change of the feature vector or weight matrix. The process involves several steps: Step S1: By comparing the patterns of the disconnector with those of surrounding stable equipment to filter out common-mode effects, the internal relative deformation information of the disconnector is determined. Step S2: Based on the dynamic characteristic model, rigid body motion or center-of-mass displacement is fitted, and the overall displacement information of the disconnector is determined by comparing it with the motion of surrounding stable equipment. Step S3: Based on the time-varying indices, an absolute attitude change information of the disconnector relative to the standard health state reference network, including absolute vertical settlement rate, pitch angle, and roll angle, is determined using a transformation model. This transformation model is a mathematical model based on coordinate transformation and geometric mechanics, used to convert the time-varying indices, such as relative distance or relative angle change rate, representing the relative geometric relationship between the monitoring point and the reference point of the standard health state reference network, output by the second analysis module, into the overall absolute attitude change information of the disconnector. Specifically, this model can use Kalman filtering or extended Kalman filtering, combined with the known structural parameters of the disconnector and the initial position of the monitoring point, to estimate its six degrees of freedom (X, Y, Z direction displacement and Roll, Pitch, Yaw angles) attitude changes in real time. For example, for absolute vertical settlement rate, the rate of change of vertical distance between monitoring point and reference point can be averaged or weighted; for pitch angle and roll angle, the relative height difference or planar projection angle of monitoring point at different positions can be solved by trigonometric geometric relationship and optimized by least squares method; the condition assessment module is used to integrate the internal relative deformation information, overall displacement information and absolute attitude change information, and generate mechanical condition assessment report including risk level assessment according to preset threshold or change trend analysis.

[0017] The verification module compares the mechanical condition assessment report with ground-measured benchmark data to generate accuracy assessment indicators. The mechanical condition assessment report includes risk level, internal relative deformation information, overall displacement information, and absolute attitude change information. Different comparison methods are used for different types of benchmark data: 1. Vibration sensor data: The internal deformation information in the assessment report, such as characteristic frequency changes and vibration amplitude, is matched with the three-dimensional acceleration time series of a high-precision vibration sensor using frequency domain or time domain features and similarity calculations. For example, the difference between the predicted fault frequency components in the assessment report and the corresponding peak values ​​in the actual vibration spectrum is calculated, or the time series similarity is compared using the Dynamic Time Warping (DTW) algorithm. 2. Inclinometer data: The absolute attitude change information in the assessment report, such as pitch angle and roll angle, is directly compared with the angle change data recorded by the inclinometer to calculate its absolute or relative error. 3. Point cloud data: The overall displacement and absolute attitude change information predicted in the assessment report are registered and analyzed with multi-period point cloud data. By comparing the rigid body transformation parameters of the disconnector switch obtained from point cloud data at different times, such as those obtained using the ICP algorithm, with the displacement and attitude prediction values ​​in the evaluation report, a difference analysis is performed. When generating accuracy evaluation indicators, the comparison results can be weighted and averaged or fused. For example, a comprehensive error function can be defined, or the results can be quantified using indicators such as F1 score or RMSE.

[0018] The system includes a calibration module, which generates calibration feedback instructions based on the accuracy evaluation index to adjust the parameters or models of at least one other module of the system through iterative feedback. The calibration module 8 generates calibration feedback instructions based on the accuracy evaluation index to adjust the parameters or models of other modules of the system through iterative feedback. When the accuracy evaluation index is lower than a preset threshold, the calibration module will initiate an adaptive calibration process: 1. Adjusting the hyperparameters or model structure of the multidimensional signal processing algorithm in the first analysis module: For example, if the evaluation results show that the model has a low recognition rate for a specific type of fault, the calibration module may suggest adjusting the learning rate, regularization parameters, or increasing the number of network layers, changing the activation function, etc., of the graph neural network. This can be achieved through hyperparameter search algorithms based on Bayesian optimization or reinforcement learning. 2. Adjusting the definition parameters of the standard health state reference network: For example, if the verification results show that the reference network fails to accurately reflect the normal behavior of the disconnector, the calibration module may fine-tune or reconstruct the reference point positions and topology in the reference network based on new ground-based measured benchmark data to better adapt to changes in the actual environment. 3. Adjust the mapping rules in the baseline state mapping module: If the mapping deviation is large, the calibration module can optimize the mapping function or weights, for example, through an iterative optimization method that minimizes the mapping error. 4. Adjust the classification thresholds used to distinguish different deformation levels in the fault interpretation module: Based on the validation results, if the false positive or false negative rate for certain fault levels is too high, the calibration module will dynamically adjust the corresponding preset thresholds, for example, through parameter optimization of classification models such as Support Vector Machine (SVM) or Logistic Regression.

[0019] When constructing the electrical parameter edge, the power source monitoring module connects the monitoring point nodes that are structurally adjacent, based on the known or extracted geometric topology of the transmission mechanism from the sensor and the disconnector switch.

[0020] The baseline state mapping module is configured to map each monitoring point node in the dynamic characteristic model to three reference points with predetermined relative geometric positions in the standard health state reference network.

[0021] The weight matrix and eigenvectors output by the first analysis module include the time-varying weight values ​​between nodes and the time-varying coordinates of the node eigenvectors.

[0022] The spatiotemporal relative geometric relationship time change index output by the second analysis module includes the time change rate or cumulative change of the relative distance or relative angle defined by the monitoring point node and the reference point set of the standard health status reference network that it maps to.

[0023] When determining internal relative deformation information, the fault interpretation module compares the time evolution pattern of the feature vector or weight matrix of the disconnecting switch with the corresponding time evolution pattern obtained from surrounding stable equipment to identify and quantify specific internal deformations indicating stress concentration or uneven settlement.

[0024] When determining the absolute attitude change information, the fault interpretation module uses a preset conversion model to calculate the absolute vertical settlement rate expressed in millimeters per year or similar units and the tilt angle change value, which includes pitch and roll angles, expressed in degrees or radians.

[0025] The ground-based benchmark data used by the verification module is selected from at least one of the following: three-dimensional acceleration time series of high-precision vibration sensors deployed on the disconnector, temperature rise change data of precision temperature measurement, angle change data recorded by inclinometer, or multi-period point cloud data obtained by ground three-dimensional scanning equipment.

[0026] The calibration feedback instruction generated by the calibration module is used to adjust at least one of the following: the model structure or hyperparameters of the multidimensional signal processing algorithm in the first analysis module, the definition parameters of the standard health state reference network, the mapping rules in the baseline state mapping module, or the classification threshold used to distinguish different deformation levels in the fault interpretation module.

[0027] The device also includes a data interface and a preprocessing module, which receives initial monitoring point detection results generated by analyzing a series of high-resolution sensor data (which includes multiphysics observation information) acquired over time and covering the disconnector switch, performs coordinate calculation, time-series matching and noise filtering, and outputs the data required for the dynamic characteristic model to the power source monitoring module.

[0028] When acquiring voltage and current data, the power source monitoring module uses high-precision current transformers and voltage divider circuits to achieve signal conversion, ensuring that the linearity and accuracy of the original signal meet the requirements of subsequent analysis. The transformer ratio is set to 1000:1, and the resistor values ​​of the voltage divider circuit are 390kΩ and 392Ω, respectively, with an accuracy of 0.1%. The final conversion relationship between the voltage signal input to the data acquisition unit and the original signal is Vi = Vin * 0.0010051.

[0029] When analyzing the dynamic characteristic model, the first analysis module uses the two-table method to calculate the total instantaneous power of the three-phase system. The formula is P_inst(t) = Vab(t) * Ia(t) + Vcb(t) * Ic(t), where Vab(t) and Vcb(t) represent the instantaneous voltages of lines AB and CB, respectively, and Ia(t) and Ic(t) represent the instantaneous currents of phase A and phase C, respectively.

[0030] When determining the internal relative deformation information, the fault interpretation module converts the time-domain signal into a frequency-domain signal through Fast Fourier Transform (FFT) to analyze the characteristic frequency components that may be excited under specific mechanical faults and the changes in their amplitudes, so as to identify faults such as bearing wear and gearbox defects.

[0031] When integrating information, the condition assessment module combines internal relative deformation information, overall displacement information, and absolute attitude change information into a multi-dimensional feature vector, and eliminates differences caused by external conditions through baseline comparison and rate of change calculation. The rate of change calculation formula is (current value - baseline value) / baseline value * 100%.

[0032] When generating calibration feedback instructions, the calibration module combines Dynamic Time Warping (DTW) and Fréchet distance algorithms to quantitatively analyze the similarity between the measured curve and the reference curve, and outputs the warping distance and minimum length as the basis for adjustment.

[0033] The beneficial effects of this invention are: 1. High data acquisition accuracy: The acquisition circuit, composed of high-precision current transformers, ensures that the measurement accuracy of the original voltage and current signals is better than 0.5%, providing a reliable data foundation for subsequent calculations.

[0034] 2. Extremely High Sensitivity to State Perception: This invention uses the instantaneous output power of the motor as the core state variable, a parameter that is extremely sensitive to changes in load torque. Any change in the mechanical state of any component in the disconnector switch drive system (such as slight bearing jamming, minor deformation of the connecting rod, or slight lubrication deficiency) will immediately cause a change in load torque, which can then be captured with high precision. Compared to traditional maintenance methods such as observing the opening and closing positions with the human eye and listening to operating sounds, this invention can detect early, subtle faults that are imperceptible to humans, achieving a leap from "macroscopic qualitative" to "microscopic quantitative."

[0035] 3. Superior selection of core state variables: The instantaneous output power method of the motor is adopted. Its curve contains more state information than a single current or angle value, which can comprehensively reflect the torque changes during operation and provide a sufficient data basis for accurate diagnosis.

[0036] 4. Multi-algorithm fusion improves recognition rate: Combining multiple algorithms such as threshold method, DTW distance, and Fréchet distance, it makes comprehensive judgments from different dimensions (amplitude, shape, statistical characteristics) and cross-verifies with each other, effectively avoiding misjudgment and missed judgment. The recognition rate of typical mechanical faults (such as jamming and asynchrony) is significantly higher than that of traditional methods.

[0037] 5. High real-time performance: The entire process from data acquisition and calculation to diagnosis can be completed within hundreds of milliseconds to several seconds after an operation is completed. It can provide real-time status assessment results, meet the needs of on-site online monitoring and instant diagnosis, and provide time guarantee for equipment status management and rapid fault response.

[0038] 6. Full coverage of equipment types: This invention is applicable to all AIS (air-insulated switchgear) and GIS (gas-insulated switchgear) disconnectors that use electric operating mechanisms, without being limited by voltage level or specific model.

[0039] 7. Full life cycle monitoring: It can be used for routine status monitoring and fault early warning of equipment in operation, as well as for performance verification of equipment after maintenance; at the same time, for newly installed disconnecting switches in newly built substations, this invention is an excellent installation, commissioning and acceptance tool, which can effectively detect potential damage to the internal mechanical structure (such as component deformation and jamming) caused by improper transportation or installation, and ensure that the initial state of the commissioned equipment is healthy.

[0040] 8. Diverse user base: Its clear and quantifiable diagnostic results can provide decision support for professionals in different roles, including dispatchers (to grasp the real-time status of equipment), maintenance personnel (to locate faults and accurately eliminate defects), installation and commissioning personnel (to ensure acceptance testing), and training personnel (to provide teaching cases and practical verification). Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the module structure of the high-voltage disconnector mechanical condition detection device of the present invention, showing the connection relationship between the power source monitoring module, the reference state mapping module, the first analysis module, the second analysis module, the fault interpretation module, the state assessment module, the verification module, and the calibration module.

[0042] Figure 2 This diagram illustrates the process of constructing the dynamic characteristic model, showing the connection relationships between time-series nodes, electrical parameter edges, and temporal edges representing the instantaneous power change of the motor, and also indicating the correspondence between nodes and the geometric topology of the transmission mechanism.

[0043] Figure 3This diagram illustrates the mapping relationship between the standard health status reference network and the dynamic characteristic model, showing how monitoring point nodes are mapped to three reference points with predetermined relative geometric positions, and marking the calculation area for the time change index of the spatiotemporal relative geometric relationship.

[0044] Figure 4 The flowchart shows the process of the fault interpretation module, including the steps for determining internal relative deformation information, overall displacement information, and absolute attitude change information, as well as the application scenarios of fast Fourier transform and characteristic frequency analysis.

[0045] Figure 5 This diagram illustrates the integrated analysis process of the state assessment module, showing how multidimensional feature vectors are generated and the specific implementation process of baseline comparison and rate of change calculation.

[0046] Figure 6 The closing waveform analysis diagram shows that for the opening and closing curves that are difficult to distinguish with the naked eye, the algorithm can calculate the quantitative indicators of significant differences, thereby achieving a state judgment with almost 100% accuracy.

[0047] The attached figures are labeled as follows: 1. Power source monitoring module; 2. Baseline state mapping module; 3. First analysis module; 4. Second analysis module; 5. Fault interpretation module; 6. State assessment module; 7. Verification module; 8. Calibration module; 9. Dynamic characteristic model; 10. Standard health state reference network. Detailed Implementation

[0048] To make the objectives, technical solutions, and beneficial effects of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so as to facilitate understanding by those skilled in the art.

[0049] This invention provides a mechanical condition detection device for high-voltage disconnect switches, the specific implementation of which is as follows. Figure 1As shown, the device includes a power source monitoring module 1, a reference state mapping module 2, a first analysis module 3, a second analysis module 4, a fault interpretation module 5, a state assessment module 6, a verification module 7, and a calibration module 8. These modules are interconnected through a data transmission interface to form a complete detection system. The power source monitoring module 1 is responsible for collecting the voltage and current signals of the drive motor and transmitting the processed dynamic characteristic model 9 to the reference state mapping module 2 and the first analysis module 3. The reference state mapping module 2 maps the nodes in the dynamic characteristic model 9 to the standard health state reference network 10, generates a mapping relationship, and outputs it to the second analysis module 4. The first analysis module 3 analyzes the dynamic characteristic model 9 and outputs the weight matrix and eigenvector to the fault interpretation module 5. The second analysis module 4 calculates the time change index of the spatiotemporal relative geometric relationship based on the mapping relationship and outputs it to the fault interpretation module 5. The fault interpretation module 5 integrates the internal relative deformation information, overall displacement information, and absolute attitude change information and transmits it to the state assessment module 6. The state assessment module 6 generates a mechanical state assessment report and transmits it to the verification module 7. The verification module 7 compares the report with the ground measured reference data and generates an accuracy assessment index and transmits it to the calibration module 8. The calibration module 8 generates calibration feedback instructions based on the accuracy assessment index to adjust the parameters or models of other modules.

[0050] In actual operation, the power source monitoring module 1 uses high-precision current transformers and voltage divider circuits to achieve signal conversion. The transformer ratio is set to 1000:1, and the resistor values ​​of the voltage divider circuit are 390kΩ and 392Ω, respectively, with an accuracy of 0.1%. The conversion relationship between the voltage signal input to the data acquisition unit and the original signal is Vi = Vin * 0.0010051. After amplification, filtering, and analog-to-digital conversion, the acquired voltage and current signals are used to construct a dynamic characteristic model 9. Figure 2 As shown, the dynamic characteristic model 9 includes time-series nodes, electrical parameter edges, and timing edges representing the instantaneous power changes of the motor. Time-series nodes correspond to critical moments in the motor's operation. Electrical parameter edges connect adjacent nodes based on the geometric topology of the disconnector switch transmission mechanism, while timing edges connect nodes within different operating cycles. For example, if node A is connected to node B via an electrical parameter edge, it indicates a direct physical connection between the two in the transmission mechanism; while if node A is connected to node C via a timing edge, it indicates that the two belong to different operating cycles but have temporal continuity.

[0051] Calculation of electrical parameter edges Geometric topological mapping: Mathematically, the weight of an edge can be quantified by equivalent mechanical impedance or energy transfer efficiency, for example: Where Zmech is the mechanical impedance and θij is the transmission angle between nodes. The calculation results need to be coupled with electrical parameters (such as current phase) to reflect the "electromechanical" coupling characteristics.

[0052] Numerical calculation example: Taking gear transmission as an example, the electrical parameter side can be represented as: Where ω is the angular velocity, Irms is the effective value of the current, and the coefficient k is determined by the gear ratio and motor parameters.

[0053] Markov chain training and implementation Training Algorithm: The training process for the Markov chain model used in the temporal edges includes: 1. State space definition: Discretize time series nodes into states (such as "high power", "low power", "abnormal fluctuation").

[0054] 2. Transition Probability Estimation: The state transition matrix P is calculated using maximum likelihood estimation (MLE), where the elements... Nij represents the number of transitions from state i to state j.

[0055] 3. Convergence optimization: Use iterative algorithms (such as Power Iteration) to solve for the stationary distribution to ensure model convergence.

[0056] Parameter optimization: Hyperparameters such as the number of states and transition order need to be adjusted through Bayesian optimization or grid search. The loss function is often chosen as the negative log-likelihood (NLL) or the BIC criterion.

[0057] Data requirements: Training requires a large amount of historical operating cycle data (recommended >10^4 sequences), and the data must cover normal and typical failure modes (such as jamming, wear).

[0058] Model integration and dynamic updates Deep learning integration: Dynamic feature models are often combined with deep learning, for example, using LSTM to encode time series nodes, using GNN to process electrical parameter edge topology, and outputting state classification probabilities.

[0059] Online learning mechanism: The model needs to support online updates, and optimize the transition probability through retraining on false positive samples (such as incremental learning of SVM) or reinforcement learning (reward function is the fault detection accuracy).

[0060] Dynamic Characteristic Model 9 is a system model requiring data-driven training. Its core lies in integrating electrical parameter data with the transmission mechanism topology, using machine learning methods to achieve deep characterization and anomaly detection of the equipment's mechanical state. Its complexity manifests in three aspects: ① Multi-source data integration: The model input includes preprocessed drive motor voltage and current data, which are then converted using high-precision current transformers (1000:1 ratio) and voltage divider circuits (resistors 390kΩ and 392Ω, accuracy 0.1%), and finally standardized according to the conversion relationship Vi=Vin×0.0010051. ② Dual association of topology and time series: The model structure includes time-series nodes (representing instantaneous power changes in the motor), electrical parameter edges (connecting adjacent monitoring points based on the transmission mechanism's geometric topology), and time-series edges (associating nodes across cycles through Markov chains or time-series analysis), forming a dynamic graph network. ③ Adaptive learning requirements: Node A is connected to node B through electrical parameter edges (physical association) and to node C through temporal edges (temporal continuity). Such associations need to be trained with historical data to quantify edge weights and state transition probabilities.

[0061] The baseline state mapping module 2 maps the nodes in the dynamic characteristic model 9 to the standard health state reference network 10. For example... Figure 3 As shown, the standard health status reference network 10 consists of multiple reference points with predetermined relative geometric positions, and each monitoring point node is mapped to three reference points. For example, node A is mapped to reference points P1, P2, and P3, where the geometric relationships between P1, P2, and P3 are known and fixed. During the mapping process, the baseline state mapping module 2 determines the mapping relationship by calculating the Euclidean distance or angular deviation between the node and the reference points and outputs it to the second analysis module 4. The second analysis module 4 calculates the temporal variation index of the spatiotemporal relative geometric relationship based on the mapping relationship, such as the rate of change or cumulative change of the relative distance or relative angle defined by the monitoring point node and its mapped set of reference points. These indices are used for subsequent fault interpretation.

[0062] The first analysis module 3 employs a multidimensional signal processing algorithm to analyze the dynamic characteristic model 9. The first analysis module 3 uses a graph neural network or time-series graph analysis algorithm to analyze the dynamic characteristic model 9. The graph neural network can be a graph convolutional network (GCN) or a graph attention network (GAT), etc. The specific implementation of this algorithm includes: first, constructing a graph data structure based on the node and edge structure of the dynamic characteristic model 9; second, defining the architecture of the graph neural network, for example, containing 2-3 graph convolutional layers, each followed by a ReLU activation function and a batch normalization layer; then, using the electrical parameter data obtained from the power source monitoring module 1 as node feature input, performing feature learning through the graph neural network, and finally outputting a weight matrix of the electrical parameter correlation between nodes and feature vectors encoding node states and neighborhood information. The training data for this network can come from a large amount of historical running data and simulation data, using, for example, a cross-entropy loss function for supervised learning, and training through the Adam optimizer. Figure 2 As shown, the analysis process includes calculating the weight matrix of the electrical parameter correlation between nodes and the feature vector encoding the node state and neighborhood information. The weight matrix contains the weight values ​​between nodes that change over time, while the feature vector contains the coordinates of the node feature vectors that change over time. For example, the weight value between node A and node B may change from 0.8 to 0.6 over time, and the coordinates of the feature vector may also change from (1, 0.5) to (0.8, 0.4). The first analysis module 3 also uses the two-table method to calculate the total instantaneous power of the three-phase system, with the formula P_inst(t) = Vab(t) * Ia(t) + Vcb(t) * Ic(t), where Vab(t) and Vcb(t) represent the instantaneous voltages of lines AB and CB, respectively, and Ia(t) and Ic(t) represent the instantaneous currents of phase A and phase C, respectively. The analysis results are output to the fault interpretation module 5.

[0063] The workflow of fault interpretation module 5 is as follows: Figure 4As shown. First, in step S1, based on the time variation pattern of the feature vector or weight matrix, the common-mode effect is filtered out by comparing it with the pattern of surrounding stable equipment to determine the internal relative deformation information of the disconnector. For example, if the time evolution pattern of the feature vector of the disconnector differs significantly from the pattern of the surrounding stable equipment, it indicates that there may be stress concentration or uneven settlement. Second, in step S2, based on the dynamic characteristic model 9, the rigid body motion is fitted or the center of mass displacement is tracked, and the overall displacement information of the disconnector is determined by comparing it with the motion of the surrounding stable equipment. For example, if the center of mass displacement velocity of the disconnector is significantly higher than that of the surrounding equipment, it indicates that there may be abnormal displacement. Finally, in step S3, based on the time variation index of the spatiotemporal relative geometric relationship, the absolute attitude change information of the disconnector relative to the standard health state reference network 10 is determined by a preset transformation model. For example, the absolute vertical settlement rate of the disconnector is calculated to be 0.5 mm / year, and the pitch angle and roll angle changes are 0.1 degrees and 0.2 degrees, respectively. The fault interpretation module 5 also converts the time-domain signal into a frequency-domain signal through fast Fourier transform, analyzes the characteristic frequency components that may be excited under specific mechanical faults and their amplitude changes, so as to identify faults such as bearing wear or gearbox defects.

[0064] The workflow of Status Assessment Module 6 is as follows: Figure 5 As shown. First, the internal relative deformation information, overall displacement information, and absolute attitude change information are combined into a multi-dimensional feature vector. For example, the feature vector may include the stress concentration index of the internal relative deformation, the velocity value of the overall displacement, and the pitch and roll angles of the absolute attitude change. Second, differences caused by external conditions are eliminated through baseline comparison and rate of change calculation. Baseline comparison refers to comparing the current value with historical reference values, and the rate of change calculation formula is (current value - reference value) / reference value * 100%. For example, if the current pitch angle is 0.1 degrees and the reference value is 0.05 degrees, then the rate of change is 100%. Finally, a mechanical condition assessment report including risk level assessment is generated based on preset thresholds or trend analysis. For example, if the pitch angle change rate exceeds 50%, the risk level is assessed as "high".

[0065] Verification module 7 compares the mechanical condition assessment report with ground-based measured benchmark data. The ground-based benchmark data is selected from at least one of the following: three-dimensional acceleration time series from high-precision vibration sensors deployed on the disconnector, temperature rise change data from precision temperature measurements, angle change data recorded by an inclinometer, or multi-period point cloud data acquired by a ground-based three-dimensional scanning device. For example, if the pitch angle change value in the mechanical condition assessment report is 0.1 degrees, while the actual change value recorded by the inclinometer is 0.09 degrees, then the accuracy assessment index generated by verification module 7 is 90%. Calibration module 8 generates calibration feedback instructions based on the accuracy assessment index, used to adjust the model structure or hyperparameters of the multi-dimensional signal processing algorithm in the first analysis module 3, the defined parameters of the standard health state reference network 10, the mapping rules in the benchmark state mapping module 2, or the classification thresholds used to distinguish different deformation levels in the fault interpretation module 5. For example, if the accuracy assessment index is below 80%, calibration module 8 may adjust the hyperparameters of the multi-dimensional signal processing algorithm to improve analytical accuracy.

[0066] In addition, this device includes a data interface and preprocessing module for receiving initial monitoring point detection results generated by analyzing a series of high-resolution sensor data covering the disconnecting switch acquired over time. The sensor data contains multiphysics observation information, such as vibration, temperature, and electromagnetic fields. After performing coordinate calculation, timing matching, and noise filtering, the data interface and preprocessing module outputs the processed data to the power source monitoring module 1. For example, if high-frequency noise exists in the sensor data, it is filtered using a low-pass filter to ensure the accuracy of subsequent analysis.

[0067] This device, through the coordinated operation of the aforementioned modules, achieves comprehensive monitoring and accurate diagnosis of the mechanical condition of high-voltage disconnect switches. For example, in a practical application, the centroid displacement velocity of the disconnect switch was detected as 0.2 mm / s, the pitch angle change was 0.15 degrees, and the characteristic frequency components showed signs of bearing wear. The condition assessment module 6 generated a mechanical condition assessment report with a risk level of "medium" based on this information. The verification module 7 confirmed the accuracy of the report by comparing it with ground-based measured data, and the calibration module 8 adjusted the hyperparameters of the multi-dimensional signal processing algorithm based on the comparison results to further improve diagnostic accuracy.

[0068] Work process: First, in the power source monitoring module 1, the voltage and current signals of the drive motor are acquired through a high-precision current transformer and a voltage divider circuit. For example... Figure 1 As shown, the current transformer converts the original high-voltage signal to a low-voltage signal with a transformation ratio of 1000:1. The resistors in the voltage divider circuit have resistance values ​​of 390kΩ and 392Ω, respectively, to ensure that the signal linearity and accuracy meet the requirements. After analog-to-digital conversion, dynamic characteristic model 9 is constructed. Figure 2As shown, the time-series nodes in this model correspond to key physical events in the motor's operation, such as the start-up, stable operation, and stopping phases. Electrical parameter edges connect adjacent nodes based on the geometric topology of the disconnector's transmission mechanism, reflecting the continuity of mechanical motion. Timing edges connect nodes within different operating cycles, demonstrating temporal correlation. For example, if node A is connected to node B via an electrical parameter edge, it indicates a direct physical relationship between the two; while if node A is connected to node C via a timing edge, it indicates that they belong to different operating cycles but have temporal continuity. This modeling approach can comprehensively capture the dynamic changes in the disconnector's operating state.

[0069] Subsequently, the baseline state mapping module 2 maps the nodes in the dynamic characteristic model 9 to the standard health state reference network 10. For example... Figure 3 As shown, the standard health status reference network 10 consists of multiple reference points, with each monitoring node mapped to three reference points at predetermined geometric locations. For example, node A is mapped to reference points P1, P2, and P3, where the geometric relationships between P1, P2, and P3 are known and fixed. During the mapping process, the mapping relationship is established by calculating the Euclidean distance or angular deviation between the node and the reference points. Based on this mapping relationship, the second analysis module 4 calculates temporal variation indicators of the spatiotemporal relative geometric relationship, such as the rate of change of the relative distance or relative angle between the monitoring node and the set of reference points it maps to over time. These indicators can quantify the geometric change trend of the disconnector during operation, providing a basis for subsequent fault interpretation.

[0070] Next, the first analysis module 3 uses a multidimensional signal processing algorithm to analyze the dynamic characteristic model 9. For example... Figure 2 As shown, the analysis process includes calculating the weight matrix of the electrical parameter correlation between nodes and the feature vector encoding the node state and neighborhood information. For example, the weight value between node A and node B may change from 0.8 to 0.6 over time, and the coordinates of the feature vector may also change from (1, 0.5) to (0.8, 0.4). Furthermore, the first analysis module 3 calculates the total instantaneous power of the three-phase system using the two-table formula P_inst(t) = Vab(t) * Ia(t) + Vcb(t) * Ic(t), where Vab(t) and Vcb(t) represent the instantaneous voltages of lines AB and CB, respectively, and Ia(t) and Ic(t) represent the instantaneous currents of phases A and C, respectively. This analysis process can extract key features of the disconnector's operating state, providing data support for subsequent fault interpretation.

[0071] In the fault interpretation module 5, step S1 is first executed. Based on the time variation pattern of the feature vector or weight matrix, common-mode effects are filtered out by comparing it with the patterns of surrounding stable equipment to determine the internal relative deformation information of the disconnector. For example, if the time evolution pattern of the feature vector of the disconnector differs significantly from the pattern of the surrounding stable equipment, stress concentration or uneven settlement may exist. Secondly, in step S2, rigid body motion is fitted or the center of mass displacement is tracked based on the dynamic characteristic model 9, and the overall displacement information of the disconnector is determined by comparing it with the motion of the surrounding stable equipment. For example, if the center of mass displacement velocity of the disconnector is significantly higher than that of the surrounding equipment, abnormal displacement may exist. Finally, in step S3, based on the time variation index of the spatiotemporal relative geometric relationship, the absolute attitude change information of the disconnector relative to the standard health state reference network 10 is determined through a preset transformation model. For example, the absolute vertical settlement rate of the disconnector is calculated to be 0.5 mm / year, and the pitch and roll angle changes are 0.1 degrees and 0.2 degrees, respectively. In addition, the fault interpretation module 5 converts the time-domain signal into a frequency-domain signal through fast Fourier transform, analyzes the characteristic frequency components that may be excited under specific mechanical faults and their amplitude changes, so as to identify faults such as bearing wear or gearbox defects.

[0072] The workflow of Status Assessment Module 6 is as follows: Figure 5 As shown. First, the internal relative deformation information, overall displacement information, and absolute attitude change information are combined into a multi-dimensional feature vector. For example, the feature vector may include the stress concentration index of the internal relative deformation, the velocity value of the overall displacement, and the pitch and roll angles of the absolute attitude change. Second, differences caused by external conditions are eliminated through baseline comparison and rate of change calculation. Baseline comparison refers to comparing the current value with historical reference values, and the rate of change calculation formula is (current value - reference value) / reference value * 100%. For example, if the current pitch angle is 0.1 degrees and the reference value is 0.05 degrees, then the rate of change is 100%. Finally, a mechanical condition assessment report including risk level assessment is generated based on preset thresholds or trend analysis. For example, if the pitch angle change rate exceeds 50%, the risk level is assessed as "high".

[0073] Verification module 7 compares the mechanical condition assessment report with ground-based measured benchmark data. The ground-based benchmark data is selected from at least one of the following: three-dimensional acceleration time series from high-precision vibration sensors deployed on the disconnector, temperature rise change data from precision temperature measurements, angle change data recorded by an inclinometer, or multi-period point cloud data acquired by a ground-based three-dimensional scanning device. For example, if the pitch angle change value in the mechanical condition assessment report is 0.1 degrees, while the actual change value recorded by the inclinometer is 0.09 degrees, then the accuracy assessment index generated by verification module 7 is 90%. Calibration module 8 generates calibration feedback instructions based on the accuracy assessment index, used to adjust the model structure or hyperparameters of the multi-dimensional signal processing algorithm in the first analysis module 3, the defined parameters of the standard health state reference network 10, the mapping rules in the benchmark state mapping module 2, or the classification thresholds used to distinguish different deformation levels in the fault interpretation module 5. For example, if the accuracy assessment index is below 80%, calibration module 8 may adjust the hyperparameters of the multi-dimensional signal processing algorithm to improve analytical accuracy.

[0074] In addition, the data interface and preprocessing module receives initial monitoring point detection results generated by analyzing a series of high-resolution sensor data acquired over time. The sensor data contains multiphysics observation information, such as vibration, temperature, and electromagnetic fields. After performing coordinate calculation, time-series matching, and noise filtering, the data interface and preprocessing module outputs the processed data to the power source monitoring module 1. For example, if high-frequency noise exists in the sensor data, it is filtered using a low-pass filter to ensure the accuracy of subsequent analysis.

[0075] Through the coordinated operation of the aforementioned modules, this device achieves comprehensive monitoring and accurate diagnosis of the mechanical condition of the high-voltage disconnector switch. For example, in a practical application, the centroid displacement velocity of the disconnector switch was detected as 0.2 mm / s, the pitch angle change was 0.15 degrees, and the characteristic frequency components showed signs of bearing wear. Based on this information, the condition assessment module 6 generated a mechanical condition assessment report with a risk level of "medium." The verification module 7 confirmed the accuracy of the report by comparing it with ground-based measured data, and the calibration module 8 adjusted the hyperparameters of the multidimensional signal processing algorithm based on the comparison results to further improve diagnostic accuracy.

[0076] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A high-voltage disconnector mechanical state detection device, characterized by, Comprise: a power source monitoring module (1) for receiving the driving motor voltage and current data processed by the preprocessing module through the data interface, constructing a dynamic characteristic model (9) of the motor operating state, which includes a time series node representing the instantaneous power change of the motor, an electrical parameter edge connecting structurally adjacent monitoring point nodes within the same operation cycle according to the known or extracted transmission mechanism geometric topology structure of the disconnecting switch, and a time sequence edge connecting nodes of different operation cycles based on time sequence analysis or Markov chain model; a reference state mapping module (2) for mapping the nodes in the dynamic characteristic model (9) to the reference points of a pre-defined standard health state reference network (10) established based on a large amount of historical health data, expert experience or simulation model, establishing a mapping relationship; a first analysis module (3) configured to analyze the dynamic characteristic model (9) using a multi-dimensional signal processing algorithm such as graph neural network or time sequence graph analysis, the graph neural network for example including at least 2 layers of graph convolution layer and taking the electrical parameter data as node feature input, outputting a weight matrix representing the correlation of electrical parameters between nodes and a feature vector encoding node state and neighborhood information; a second analysis module (4) configured to process the mapping relationship and output a time-varying index representing the spatio-temporal relative geometric relationship of the monitoring points with respect to the reference points of the standard health state reference network (10) through geometric analysis; a fault interpretation module (5) for performing the following steps: step S1, determining the internal relative deformation information of the disconnecting switch by comparing the mode with the surrounding stable equipment to filter the common mode effect based on the time-varying mode of the feature vector or weight matrix; step S2, determining the overall displacement information of the disconnecting switch by comparing the motion with the surrounding stable equipment based on the fitting of rigid motion or tracking of the dynamic characteristic model (9); step S3, determining the absolute attitude change information of the disconnecting switch with respect to the standard health state reference network (10) including absolute vertical settlement rate, pitch angle and roll angle by combining the conversion model of Kalman filter or extended Kalman filter and combining the known structural parameters of the disconnecting switch and the initial position of the monitoring points based on the time-varying index; a verification module (7) for comparing the mechanical state evaluation report with the ground measured reference data to generate an accuracy evaluation index; and a calibration module (8) for generating calibration feedback instructions according to the accuracy evaluation index to adjust the parameters or models of at least one other module of the system through iterative feedback.

2. The mechanical state detection device for high-voltage disconnectors according to claim 1, characterized in that: When constructing the electrical parameter edge, the power source monitoring module (1) connects the structurally adjacent monitoring point nodes according to the known or extracted transmission mechanism geometric topology structure of the disconnecting switch.

3. The mechanical state detection device for high-voltage disconnectors according to claim 1 or 2, characterized in that: The reference state mapping module (2) is configured to map each monitoring point node in the dynamic characteristic model (9) to three reference points with predetermined relative geometric positions in the standard health state reference network (10).

4. The mechanical state detection device for high-voltage disconnectors according to claim 1, characterized in that: The weight matrix and eigenvector output by the first analysis module (3) include time-varying node weight values and time-varying node eigenvector coordinates.

5. The mechanical state detection device for high-voltage disconnectors according to claim 1 or 2, characterized in that: The time-varying index of spatial-temporal relative geometric relationship output by the second analysis module (4) includes the time-varying rate or cumulative change of the relative distance or relative angle defined by the monitoring point node and the reference point set of the standard health state reference network (10) mapped by the monitoring point node.

6. The mechanical state detection device for high-voltage disconnectors according to claim 5, characterized in that: The fault interpretation module (5) compares the time evolution pattern of the eigenvector or weight matrix of the disconnecting switch with the corresponding time evolution pattern obtained from the surrounding stable equipment to identify and quantify specific internal deformation indicating stress concentration or uneven settlement when determining internal relative deformation information.

7. The mechanical state detection device for high-voltage disconnectors according to claim 6, characterized in that: The fault interpretation module (5) calculates the absolute vertical subsidence rate represented by millimeters per year and the tilt angle change value represented by degrees including pitch angle and roll angle through a preset conversion model from the time-varying index of spatial-temporal relative geometric relationship when determining absolute attitude change information.

8. The mechanical state detection device for high-voltage disconnectors according to claim 1, characterized in that: The ground-based benchmark data used by the verification module (7) is selected from at least one of the following: three-dimensional acceleration time series of high-precision vibration sensors arranged on the disconnecting switch, temperature rise change data of precise temperature measurement, angle change data recorded by an inclinometer, or multi-period point cloud data obtained by a ground three-dimensional scanning device.

9. The mechanical state detection device for high voltage disconnector according to any one of claims 1-8, characterized in that: The calibration feedback instructions generated by the calibration module (8) are used to adjust at least one of the following: the model structure or hyperparameters of the multi-dimensional signal processing algorithm in the first analysis module (3), the definition parameters of the standard health state reference network (10), the mapping rules in the benchmark state mapping module (2), or the classification threshold values in the fault interpretation module (5) for distinguishing different deformation levels.

10. The mechanical state detection device for high-voltage disconnectors according to claim 1, characterized in that: A data interface and pre-processing module are also included, which is used to receive initial monitoring point detection results generated by analyzing a series of time-varying high-resolution sensor data covering the disconnecting switch, and perform coordinate calculation, time sequence matching and noise filtering, and output the required data for the power source monitoring module (1) to the dynamic characteristic model (9).

Images