A method and system for monitoring the state of a switch component

Abstract

The application discloses a kind of switch component state monitoring method and system, it is related to electric power technical field, the method includes: from the electric parameter of switch component of distribution control equipment, extract single signal feature, and carry out feature fusion and dimension reduction, obtain switch component comprehensive feature;Using random forest algorithm, combined with the history data of switch component comprehensive feature and expert calibration, mapping from switch component comprehensive feature to health index is established;Using time series prediction model, the real-time change trend of health index is monitored, and after reaching preset condition, trigger abnormal alarm;Based on switch component comprehensive feature and health index, combined with the degradation trajectory quantitative value of switch component and dynamic degradation quantity threshold, the remaining life prediction result of switch component is output.The application improves the reliability of remaining life prediction in switch component state monitoring.

Description

Technical Field

[0001] This invention relates to the field of power technology, and more specifically, to a method and system for monitoring the status of switching components. Background Technology

[0002] In power systems, the switching components of distribution control equipment are responsible for opening and closing circuits and protecting hardware devices (such as circuit breakers and disconnectors). Their operating status can be reflected by indicators such as contact pressure and arc energy. Monitoring the status of these switching components involves data acquisition, signal processing, health assessment, and fault early warning to ensure the safety and stability of the equipment during operation.

[0003] Patent application number 202111675600.4 discloses a method and device for monitoring the status of switch components in power distribution room equipment based on deep learning. Through image recognition, ledger text parsing, and status comparison based on deep learning, it not only achieves accurate monitoring of equipment status but also generates timely alarms when anomalies are detected. However, the aforementioned method and device for monitoring the status of switch components in power distribution room equipment based on deep learning still cannot solve the following shortcomings of the prior art:

[0004] (1) Under the noise interference and nonlinear degradation mechanism, the existing technology has the problems of difficulty in capturing the real degradation trend, inability to accurately quantify the health status of equipment and inaccurate estimation of remaining life.

[0005] (2) In the prior art, the static threshold cannot respond in time when the operating status of the switching component changes dynamically, and the maintenance strategy is difficult to adapt to the actual situation, resulting in the failure of the early warning or excessive maintenance cost.

[0006] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0007] In view of the problems in the related technologies, the present invention proposes a method and system for monitoring the status of switching components, so as to overcome the above-mentioned technical problems existing in the existing related technologies.

[0008] Therefore, the specific technical solution adopted by the present invention is as follows:

[0009] According to one aspect of the present invention, a method for monitoring the state of a switching component is provided, comprising:

[0010] S1. Extract single-signal features from the electrical parameters of the switching components of the power distribution control equipment, and perform feature fusion and dimensionality reduction to obtain the comprehensive features of the switching components;

[0011] S2. Using the random forest algorithm, combined with historical data of the comprehensive characteristics of the switching components and expert calibration, a mapping from the comprehensive characteristics of the switching components to the health index is established; using a time series prediction model, the real-time trend of the health index is monitored, and an abnormal alarm is triggered when the preset conditions are reached.

[0012] S3. Based on the comprehensive characteristics and health index of the switching components, and combined with the quantified value of the degradation trajectory and the dynamic degradation threshold of the switching components, output the prediction result of the remaining life of the switching components, specifically including:

[0013] A physical damage model and a data-driven model were constructed for the switching components, and the models were fused by adaptive weights to obtain the comprehensive degradation trajectory.

[0014] A dynamic degradation threshold is constructed, and the remaining life of the switching component is calculated by combining the comprehensive degradation trajectory.

[0015] The comprehensive characteristics of the switching components include the contact pressure and arc energy within the switching components.

[0016] Furthermore, single-signal features are extracted from the electrical parameters of the switching components of the power distribution control equipment, and feature fusion and dimensionality reduction are performed to obtain the comprehensive features of the switching components, including:

[0017] The contact pressure signal and arc energy signal of the switching component are acquired, and feature extraction and feature fusion are performed on the contact pressure signal and arc energy signal.

[0018] Principal component analysis was used to reduce the dimensionality of the fused features, resulting in the comprehensive features of the switching components.

[0019] Furthermore, using the random forest algorithm, combined with historical data and expert calibration of the comprehensive characteristics of switching components, a mapping from the comprehensive characteristics of switching components to the health index is established, including:

[0020] By adding an expert-calibrated health index to the historical data of the comprehensive characteristics of the switching components, a dataset is obtained, which is then divided into a training set, a validation set, and a test set.

[0021] Using the training set and based on the preset objective function, the random forest algorithm is used to fit the mapping from the comprehensive features of the switch components to the health index.

[0022] The Analytic Hierarchy Process (AHP) is used to determine the weights of different features in the comprehensive features of the switching component, which will serve as the basis for feature selection in the Random Forest algorithm.

[0023] During training, the optimal parameters of the random forest model are selected through cross-validation; the random forest model is validated and its performance is evaluated using the validation and test sets; the comprehensive features of the switching components acquired in real time are input into the random forest model, and the predicted health index is output.

[0024] Furthermore, the analytic hierarchy process (AHP) is used to determine the weights of different features in the comprehensive features of the switching component, which serve as the basis for feature selection in the random forest algorithm.

[0025] The mapping from the comprehensive characteristics of switching components to the health index is defined as the target layer, the factors that affect the achievement of the target are defined as the criterion layer, and the characteristic variables are defined as the scheme layer.

[0026] For each element in each layer, a judgment matrix is ​​constructed using a pairwise comparison method; the maximum eigenvalue and the corresponding eigenvector of each judgment matrix are calculated; the eigenvector represents the importance weight of each element relative to a certain element in the previous layer.

[0027] The weights of each level are combined to obtain the total ranking weight of the scheme layer relative to the target layer; based on the total ranking weight, the features in the random forest algorithm are filtered.

[0028] Furthermore, by utilizing time-series prediction models, the real-time trends of health indices are monitored, and abnormal alarms are triggered upon reaching preset conditions, including:

[0029] A dynamic model of the health index is constructed using a time-series prediction model to capture the time-series variation patterns of the health index.

[0030] The system uses a dynamic model of the health index to predict values ​​for future moments; it also pre-sets the upper and lower bounds of the health index for the switching components.

[0031] If the health index of a switch component is greater than the upper limit or less than the lower limit at a certain future moment, it is predicted that the switch component will be in an abnormal state at that future moment, and the alarm module will be invoked to send an alarm signal to the monitoring center.

[0032] Furthermore, a physical damage model and a data-driven model were constructed for the switching components, and adaptive weights were used for model fusion to obtain the comprehensive degradation trajectory, including:

[0033] Based on Holm's contact law, the cumulative damage to contact resistance caused by arc energy is calculated, and a physical damage model is constructed. The expression for the physical damage model is as follows:

[0034] ;

[0035] In the formula, Δ R ( tThis represents the cumulative damage to the contact resistance caused by the electric arc energy.

[0036] k Indicates the material coefficient. t Represents a time variable;

[0037] Using a self-attention neural network, the temporal relationship between the comprehensive characteristics and health index of the switching component is analyzed, and a data-driven model of the switching component is output. The expression of the data-driven model is:

[0038] ;

[0039] In the formula, D 1( t ) represents a data-driven nonlinear degradation term;

[0040] f This represents a neural network function based on a self-attention mechanism;

[0041] X ( t This indicates the overall characteristics of the switching component. HI ( t () indicates the health index. t Represents a time variable;

[0042] Adaptive weights are determined based on the noise level of the electrical parameters of the switching components;

[0043] The physical damage model and the data-driven model of the switching component are fused according to the adaptive weights to obtain the comprehensive degradation trajectory, which is used to represent the quantitative value of the current cumulative damage of the switching component.

[0044] Furthermore, the comprehensive degradation trajectory includes:

[0045] In the physical damage model of the switching component, the average growth rate of the cumulative damage value over time is used as the drift coefficient corresponding to the physical degradation term.

[0046] By subtracting the degradation data actually measured by the switching component from the damage accumulation trend constructed by the physical damage model, several error terms are obtained, and the standard deviation of the error terms is calculated as the diffusion coefficient corresponding to the physical degradation term.

[0047] In the data-driven model of the switching component, the time-series dependency is extracted from the comprehensive features and health index of the switching component using a self-attention mechanism neural network, and the data-driven nonlinear degradation term is output. By analyzing the trend of the data-driven nonlinear degradation term, the drift coefficient corresponding to the data-driven model is obtained.

[0048] The noise and nonlinear fluctuations inherent in the comprehensive features and health index of the switching components are captured by a self-attention mechanism neural network, and the diffusion coefficient of the data-driven part is generated.

[0049] The drift coefficient and diffusion coefficient of the fusion are obtained by combining the drift coefficient and diffusion coefficient of the physical degradation term and the data-driven part with the weighted summation of the adaptive weights.

[0050] The expression for the comprehensive degradation trajectory is:

[0051] ;

[0052] In the formula, Y ( T ) indicates in T The overall degradation status of the switching components at all times. Y (0) indicates the initial degradation state of the switching component;

[0053] λ ( s ) represents the drift coefficient after fusion. σ ( s () represents the diffusion coefficient after fusion;

[0054] dW ( s ) represents standard Brownian motion W ( s At that moment s The increment.

[0055] Furthermore, a dynamic degradation threshold is constructed, and combined with the comprehensive degradation trajectory, the remaining lifetime of the switching component is calculated, including:

[0056] Identify the game participants, including maintenance decision-makers, equipment users, operations and maintenance teams, and production teams; configure game strategies, including strategies favoring early maintenance and strategies favoring delayed maintenance; determine cost parameters, including costs of early maintenance, costs of over-maintenance, failure losses, and costs of delayed maintenance risks.

[0057] Construct payoff matrices based on different game strategies and cost parameters, including conservative payoffs and aggressive payoffs;

[0058] The dynamic evolution equation of evolutionary game theory is used to calculate the change in strategy proportion, and the initial dynamic degradation threshold is determined based on the Nash equilibrium point.

[0059] Based on the historical maintenance records of the switch components, the urgency scores of different game participants were statistically analyzed;

[0060] Calculate the consensus level of each participant in the current game, using the following formula:

[0061] ;

[0062] In the formula, w i This represents the weight of each participant in the game. U ( t ) represents the time when each participant in the game is at time . t Urgency score, i Indexes representing the participants in the game;

[0063] By amplifying the final game threshold when consensus is high and keeping the final game threshold stable when consensus is low, the initial dynamic degradation threshold is adjusted, and the dynamic degradation threshold is finally obtained.

[0064] Analyze the comprehensive degradation trajectory of the switching component. When the quantified value of the current cumulative damage of the switching component in the comprehensive degradation trajectory is equal to the dynamic degradation threshold, determine that moment as the failure moment of the switching component and obtain the remaining life of the switching component.

[0065] Furthermore, based on the historical maintenance records of the switch components, the urgency scores of different game participants were statistically analyzed, including:

[0066] Collect historical maintenance records for switch components, including maintenance dates, maintenance types, and descriptions of problems found;

[0067] Identify historical maintenance records directly relevant to each game participant and set urgency scoring criteria for different types of issues and maintenance activities;

[0068] For each maintenance record, an urgency score is assigned to that maintenance record based on the urgency scoring criteria;

[0069] For each game participant, the urgency scores corresponding to all maintenance records directly related to the game participant are summarized to obtain the game participant's urgency score.

[0070] According to another aspect of the present invention, a switch component status monitoring system is also provided, including a switch component feature acquisition module, a health index acquisition and anomaly early warning module, and a lifespan prediction module.

[0071] The switch component feature acquisition module extracts single-signal features from the electrical parameters of the switch components of the power distribution control equipment, performs feature fusion and dimensionality reduction to obtain the comprehensive features of the switch components; the health index acquisition and anomaly early warning module uses a random forest algorithm, combined with historical data and expert calibration of the comprehensive features of the switch components, to establish a mapping from the comprehensive features of the switch components to the health index; it uses a time series prediction model to monitor the real-time trend of the health index and triggers an anomaly alarm when preset conditions are reached; the lifespan prediction module outputs the remaining lifespan prediction results of the switch components based on the comprehensive features and health index of the switch components, combined with the quantized value of the degradation trajectory of the switch components and the dynamic degradation threshold.

[0072] The beneficial effects of this invention are as follows:

[0073] (1) This invention utilizes random forest to establish a mapping between comprehensive features and health index, and combines time series prediction to achieve accurate quantification and trend monitoring of health status, solving the problems of difficulty in capturing the real degradation trend and difficulty in quantifying health status under nonlinear degradation mechanism; through the adaptive fusion of physical damage model and data-driven model to obtain comprehensive degradation trajectory, and based on evolutionary game theory and team consensus mechanism to construct dynamic degradation threshold, which not only improves the accuracy of remaining life estimation, but also enables the threshold to dynamically respond to changes in the operating status of switching components, avoiding early warning failure or excessive maintenance costs caused by static threshold.

[0074] (2) This invention extracts single-signal features, fuses features and reduces dimensions of electrical parameters of switching components, and then combines random forest to establish a health index mapping and time series prediction model to monitor the dynamic changes of health index, which effectively improves the detection accuracy of fault status and the accuracy of early warning, and ensures that abnormal situations can be detected in time in the future.

[0075] (3) This invention integrates a physical damage model based on Holm's contact law with a data-driven (self-attention mechanism neural network) model, and uses adaptive weights to perform a weighted summation of the drift coefficient and diffusion coefficient, accurately quantifying the cumulative damage (i.e., the comprehensive degradation trajectory) of the switching components. This provides an accurate basis for predicting remaining life and helps to achieve a scientific and reasonable maintenance strategy.

[0076] (4) This invention utilizes evolutionary game theory and team consensus mechanism to construct a dynamic degradation threshold. This threshold can be automatically adjusted according to real-time data and the urgency scores of each participant, thereby realizing intelligent early warning and timely alarm for abnormal states, and providing quantitative support for subsequent maintenance decisions. Attached Figure Description

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0078] Figure 1 This is a flowchart of a method for monitoring the status of a switching component according to an embodiment of the present invention;

[0079] Figure 2 This is a block diagram of a switch component status monitoring system according to an embodiment of the present invention.

[0080] In the picture:

[0081] 1. Switch component feature acquisition module; 2. Health index acquisition and abnormal early warning module; 3. Lifespan prediction module. Detailed Implementation

[0082] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0083] According to an embodiment of the present invention, a method and system for monitoring the status of switching components are provided.

[0084] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for monitoring the state of a switching component is provided, comprising:

[0085] S1. Extract single-signal features from the electrical parameters of the switching components of the power distribution control equipment, and perform feature fusion and dimensionality reduction to obtain the comprehensive features of the switching components. The comprehensive features of the switching components include the contact pressure and arc energy in the switching components.

[0086] In one embodiment, single-signal features are extracted from the electrical parameters of the switching components of the power distribution control equipment, and feature fusion and dimensionality reduction are performed to obtain comprehensive features of the switching components, including:

[0087] The contact pressure signal and arc energy signal of the switching component are acquired, and feature extraction and feature fusion are performed on the contact pressure signal and arc energy signal. Principal component analysis is used to reduce the dimensionality of the fused features to obtain the comprehensive features of the switching component.

[0088] For the switching components of power distribution control equipment, the main electrical parameters collected include contact pressure signals and arc energy signals, both of which are continuous signals that vary over time. Statistical information and dynamic features reflecting the equipment status are extracted from each signal to reduce redundant information in subsequent processing and capture key health information. For contact pressure and arc energy signals, the mean and standard deviation are extracted. Vector concatenation is used for feature fusion. Dimensionality reduction methods are employed to extract the most important feature components, reducing model complexity and suppressing the influence of noise and redundant information. Principal component analysis is a common dimensionality reduction algorithm.

[0089] S2. Using the random forest algorithm, combined with historical data of the comprehensive characteristics of the switching components and expert calibration, a mapping from the comprehensive characteristics of the switching components to the health index is established; using a time series prediction model, the real-time trend of the health index is monitored, and an abnormal alarm is triggered when the preset conditions are reached.

[0090] In one embodiment, using the random forest algorithm, combined with historical data of the comprehensive characteristics of switching components and expert calibration, a mapping from the comprehensive characteristics of switching components to a health index is established, including:

[0091] A dataset is obtained by adding an expert-calibrated health index to historical data of the comprehensive features of switch components, and then dividing the dataset into training, validation, and test sets. Using the training set, and based on a predefined objective function, a random forest algorithm is employed to fit the mapping from the comprehensive features of switch components to the health index. The analytic hierarchy process (AHP) is used to determine the weights of different features in the comprehensive features of switch components, serving as the basis for feature selection in the random forest algorithm. During training, cross-validation is used to select the optimal parameters for the random forest model; the validation and test sets are used to validate and evaluate the performance of the random forest model; and the real-time acquired comprehensive features of switch components are input into the random forest model, which then outputs the predicted health index.

[0092] In one embodiment, the analytic hierarchy process (AHP) is used to determine the weights of different features in the overall features of the switching component, which serve as the basis for feature selection in the random forest algorithm.

[0093] The algorithm establishes a mapping from the comprehensive characteristics of switching components to the health index as the target layer, the main factors influencing the achievement of the target as the criterion layer, and the feature variables as the scheme layer. For each element in each layer, a judgment matrix is ​​constructed using a pairwise comparison method; the maximum feature and corresponding eigenvector of each judgment matrix are calculated; the eigenvector represents the importance weight of each element relative to an element in the previous layer. The weights of each layer are combined to obtain the total ranking weight of the scheme layer relative to the target layer; based on the total ranking weight, the features in the random forest algorithm are selected.

[0094] This invention acquires a large amount of comprehensive feature data of switching components from historical data. These features typically originate from previously extracted and fused signals (such as contact pressure statistics, arc energy statistics, etc.). Corresponding expert-calibrated indices are added as labels to the dataset, thus each data sample contains both input features and an output label (health index).

[0095] Random forests are an ensemble learning method that improves the robustness and resistance to overfitting of a model by constructing multiple decision trees and voting on or averaging their outputs. This study uses a random forest model to map the comprehensive features of switch components to a health index, aiming to find a functional relationship. The objective function is to minimize the error between the predicted health index and the expert-calibrated health index, using the mean squared error as the objective function.

[0096] A mapping from the comprehensive characteristics of switching components to a health index is established as the target layer; the main factors affecting the mapping effect (such as the ability of each characteristic to reflect the physical state of the equipment, noise level, dynamic response, etc.) are designated as the criterion layer; and the actual characteristics (such as the mean and standard deviation of contact pressure, and relevant indicators of arc energy) are designated as the scheme layer. For elements within the criterion and scheme layers, a pairwise comparison method is used to construct a judgment matrix. For any two elements, an evaluation value is assigned based on their importance to the upper-level target. The weight of each element is determined by calculating the maximum eigenvalue of the judgment matrix and the corresponding eigenvector (after normalization).

[0097] The weights obtained from each level are combined to obtain the final total ranking weight of each feature relative to the target layer. Based on these weights, the most critical features can be selected from the original feature set, and the sampling priority of features can be set or the splitting process of the model can be adjusted by using weights during the training of the random forest. By using the analytic hierarchy process, the random forest model pays more attention to features that have a strong ability to express the health status and have good resistance to interference when considering each feature.

[0098] In one embodiment, using a time-series prediction model to monitor the real-time trend of health indices and triggering an anomaly alarm upon reaching preset conditions includes:

[0099] A dynamic model of the health index is constructed using a time-series prediction model to capture the temporal variation pattern of the health index. This dynamic model predicts future values ​​for the health index; upper and lower bounds are pre-set for the health index of the switching components. If, at a future moment, the health index of the switching component is greater than the upper bound or less than the lower bound, it is predicted that the switching component is in an abnormal state at that future moment, and the alarm module is invoked to send an alarm signal to the monitoring center.

[0100] The time-series prediction model employs a common, well-fitted statistical model (such as ARIMA), and the fitting process follows the standard ARIMA model fitting procedure. Based on historical data and expert experience, upper and lower bounds for the health index are set. Once the predicted health index for a future time period is detected to exceed the safety threshold, the system immediately triggers an alarm mechanism, outputting an abnormal alarm signal to the monitoring center for timely intervention. The ARIMA model can make reasonable predictions about future health indices based on historical data, detect potential anomalies in advance, and has a relatively low computational load, making it suitable for real-time monitoring system applications.

[0101] S3. Based on the comprehensive characteristics and health index of the switching components, and combined with the quantified value of the degradation trajectory and the dynamic degradation threshold of the switching components, the remaining life prediction result of the switching components is output.

[0102] In one embodiment, based on the comprehensive characteristics and health index of the switching components, and combined with the quantified value of the degradation trajectory and the dynamic degradation threshold of the switching components, the output of the remaining lifetime prediction result of the switching components includes:

[0103] A physical damage model and a data-driven model were constructed for the switching component, and the models were fused using adaptive weights to obtain a comprehensive degradation trajectory. Based on evolutionary game theory and a team consensus mechanism, a dynamic degradation threshold was constructed, and the remaining lifespan of the switching component was calculated by combining the comprehensive degradation trajectory.

[0104] In one embodiment, a physical damage model and a data-driven model are constructed for the switching component, and the models are fused using adaptive weights to obtain a comprehensive degradation trajectory, including:

[0105] Based on Holm's contact law, the cumulative damage to contact resistance caused by arc energy is calculated, and a physical damage model (i.e., physical degradation term) is constructed. The expression for the physical damage model is as follows:

[0106] ;

[0107] In the formula, Δ R ( t This represents the cumulative damage to the contact resistance caused by the electric arc energy. k This represents the material coefficient, calibrated through historical experiments. t Represents a time variable.

[0108] Using a self-attention neural network, the temporal relationship between the comprehensive characteristics and health index of the switching components is analyzed, and a data-driven model of the switching components (i.e., a data-driven nonlinear degradation term) is output. The expression of the data-driven model is:

[0109] ;

[0110] In the formula,D 1( t ) represents a data-driven nonlinear degradation term, and represents the degradation trend value captured by the neural network; f This represents a neural network function based on a self-attention mechanism, used to extract temporal dependency information from the input data; X ( t This indicates the overall characteristics of the switching component. HI ( t () indicates the health index. t Represents a time variable.

[0111] Furthermore, self-attention mechanisms are capable of capturing long-range dependencies in sequences, making them particularly suitable for processing time-series data. They allow models to focus on specific parts of the input sequence, thereby gaining a better understanding of the important information within the sequence.

[0112] Switching components in time t The comprehensive characteristics are the result of extracting and fusing the contact pressure signal and the arc energy signal. Specifically: Contact pressure signal: including maximum, minimum, and average values. Arc energy signal: including maximum, minimum, and average values.

[0113] The adaptive weights are determined based on the noise level of the electrical parameters of the switching components. The drift coefficients and diffusion coefficients of the physical damage model and the data-driven model of the switching components are fused according to the adaptive weights to obtain the comprehensive degradation trajectory, which is used to represent the quantitative value of the current cumulative damage of the switching components.

[0114] In one embodiment, obtaining the comprehensive degradation trajectory includes:

[0115] In the physical damage model of the switching component, the average growth rate of the cumulative damage over time is used as the drift coefficient corresponding to the physical degradation term. Several error terms are obtained by subtracting the actual measured degradation data of the switching component from the damage accumulation trend constructed by the physical damage model, and the standard deviation of these error terms is calculated as the diffusion coefficient corresponding to the physical degradation term. In the data-driven model of the switching component, a self-attention mechanism neural network is used to extract the temporal dependence from the comprehensive features and health index of the switching component, outputting a data-driven nonlinear degradation term (capturing complex degradation patterns and dynamic changes). By analyzing the trend of the data-driven nonlinear degradation term, the corresponding drift coefficient is obtained. The self-attention mechanism neural network captures the inherent noise and nonlinear fluctuations in the comprehensive features and health index of the switching component, generating the diffusion coefficient of the data-driven part. Based on the physical degradation term and the drift coefficient and diffusion coefficient of the data-driven part, and combined with the weighted summation of adaptive weights, the fused drift coefficient and fused diffusion coefficient are obtained.

[0116] The expression for the comprehensive degradation trajectory is:

[0117] ;

[0118] In the formula, Y ( T ) indicates in T The overall degradation status or cumulative damage of the switching components at all times. Y (0) indicates the initial degradation state of the switching component; λ ( s The ) represents the drift coefficient after fusion, used to describe the average growth rate or degradation trend during equipment degradation. σ ( s () represents the diffusion coefficient after fusion, used to describe the random fluctuations and uncertainties in the degradation process; dW ( s ) represents standard Brownian motion W ( s At that moment s The increment is used to characterize the random perturbations in the degradation process.

[0119] Holm's contact law describes resistance changes in electrical contacts, particularly the effect of arc energy on contact resistance. In electrical contacts, contact resistance changes are primarily influenced by contact pressure and arc energy. Contact pressure determines the number and quality of contact points, while arc energy damages these contact points, leading to increased resistance. The purpose of data-driven analysis is to capture the temporal correlations in the comprehensive characteristics and health indices of switching components using data-driven methods, thereby obtaining nonlinear degradation trends.

[0120] Furthermore, when determining the weights of the physical model and the data-driven model based on the noise level of the measured data, these two weights are summed to one. The two drift coefficients and two diffusion coefficients are each summed according to their respective weights to obtain the fused drift coefficient and the fused diffusion coefficient, respectively. The drift coefficient represents the deterministic average trend during the degradation process of the switching component, reflecting the rate of damage accumulation. The diffusion coefficient describes the fluctuations caused by random disturbances and uncertainties, reflecting the irregularity of the degradation process. In the integrated degradation trajectory, these two coefficients are effectively fused through adaptive weights, preserving determinism while considering random fluctuations, providing a mathematical foundation and model guarantee for accurate prediction of remaining lifetime.

[0121] In one embodiment, based on evolutionary game theory and team consensus mechanisms, a dynamic degradation threshold is constructed, and combined with the comprehensive degradation trajectory, the remaining lifetime of the switching component is calculated, including:

[0122] Identify the game participants, including maintenance decision-makers, equipment users, operations and maintenance teams, and production teams; configure game strategies, including those favoring early maintenance and those favoring delayed maintenance; determine cost parameters, including costs of early maintenance, over-maintenance, failure losses, and the risk costs of delayed maintenance. Construct payoff matrices based on different game strategies and cost parameters, including conservative and aggressive payoffs. Calculate the changes in strategy proportions using dynamic evolution equations from evolutionary game theory, and determine the initial dynamic degradation threshold based on the Nash equilibrium point. Based on historical maintenance records of the switching components, statistically analyze the urgency scores of different game participants; calculate the current consensus level of each game participant using the following formula:

[0123] ;

[0124] In the formula, w i This represents the weight of each participant in the game. U ( t ) represents the time when each participant in the game is at time . t Urgency score, i This represents the index of each participant in the game.

[0125] By amplifying the final game threshold when consensus is high and maintaining its stability when consensus is low, the initial dynamic degradation threshold is adjusted, ultimately yielding the dynamic degradation amount threshold. The comprehensive degradation trajectory of the switching component is analyzed. When the quantified value of the current accumulated damage of the switching component in the comprehensive degradation trajectory equals the dynamic degradation amount threshold, this moment is determined as the failure moment of the switching component, and the remaining lifespan of the switching component is obtained.

[0126] In one embodiment, based on the historical maintenance records of the switching component, the urgency scores of different game participants are calculated as follows:

[0127] Collect historical maintenance records for switch components, including maintenance date, maintenance type, and a description of the problem found. Identify historical maintenance records directly relevant to each game participant and establish urgency scoring criteria for different types of problems and maintenance activities. For each maintenance record, assign an urgency score based on the urgency scoring criteria.

[0128] For each game participant, the urgency scores corresponding to all maintenance records directly related to the game participant are summarized to obtain the game participant's urgency score.

[0129] The game participants include: Maintenance decision-makers: responsible for the final maintenance decision, focusing on preventative maintenance and cost control. Equipment users: concerned with the continuity and safety of switchgear usage. Operations and maintenance team: assessing the current condition of switchgear based on on-site inspection data and maintenance experience. Production team: concerned with production plans and economic losses due to switchgear downtime. A strategy favoring early maintenance advocates for early maintenance of switchgear to reduce the risk of failure and accident losses, despite incurring additional maintenance costs; a strategy favoring delayed maintenance aims to save short-term costs while facing high losses due to equipment failure. The dynamic evolution equation in evolutionary game theory is used to calculate the change in strategy proportions, which takes the form:

[0130] ;

[0131] In the formula, z m Indicates the selection strategy m The percentage of participants (e.g., those who perform maintenance earlier or later); indicating the strategy adopted. m Expected returns This represents the weighted average of all strategies. Representation Strategy m The rate of change of the proportions. After a period of dynamic evolution, the proportions of each strategy tend to stabilize, that is, reaching Nash equilibrium. At this point, based on the cost-benefit balance corresponding to each strategy, a preliminary dynamic degradation threshold can be determined. This threshold reflects the critical state before the switching component degrades to failure under the balance of interests among multiple parties.

[0132] Maintenance record collection: Collect historical maintenance records of switch components according to maintenance date, maintenance type, and problem description, and associate these records with the management responsibilities of each game participant. Establish urgency scoring criteria: Develop detailed urgency scoring criteria for different types of problems and maintenance activities (e.g., safety hazards, production interruptions, equipment aging warnings, etc.). Assign an urgency score to each record according to the criteria. For each game participant, select maintenance records directly related to them and summarize the urgency scores of each record to obtain an overall score. Weighted sum the urgency scores of each participant according to their weights to obtain the consensus level at the current moment. When the consensus level is high, it indicates that most participants believe the equipment condition is urgent. In this case, the initial dynamic degradation threshold should be appropriately increased (adjusted) to reflect the need for early maintenance or more sensitive warnings. When the consensus level is low, the initial dynamic degradation threshold should be kept stable to avoid premature or misjudgment of maintenance needs.

[0133] like Figure 2 As shown, according to another embodiment of the present invention, a switching component status monitoring system is also provided, comprising:

[0134] The switch component feature acquisition module 1 is used to extract single signal features from the electrical parameters of the switch components of the power distribution control equipment, and perform feature fusion and dimensionality reduction to obtain the comprehensive features of the switch components.

[0135] The Health Index Acquisition and Anomaly Warning Module 2 is used to establish a mapping from the comprehensive characteristics of the switch components to the health index by using the random forest algorithm, combined with historical data and expert calibration of the comprehensive characteristics of the switch components; it uses a time series prediction model to monitor the real-time trend of the health index and triggers an anomaly alarm when the preset conditions are reached.

[0136] The lifespan prediction module 3 is used to output the remaining lifespan prediction result of the switching components based on the comprehensive characteristics and health index of the switching components, combined with the quantified value of the degradation trajectory of the switching components and the dynamic degradation threshold.

[0137] To facilitate understanding of the above technical solutions of the present invention, the working principle of the present invention in actual process will be described in detail below.

[0138] I. Feature Extraction and Dimensionality Reduction:

[0139] For the contact pressure signal, samples were taken once per second over 24 hours, resulting in 86,400 sample points. The maximum value of these sample points was calculated to be 500 N, the minimum value to be 200 N, the mean to be 350 N, and the standard deviation to be 60 N. Similarly, for the arc energy signal, samples were taken over the same time period, resulting in a maximum value to be 120 J, a minimum value to be 40 J, a mean to be 80 J, and a standard deviation to be 25 J.

[0140] In addition to the basic statistics mentioned above, advanced statistical features such as kurtosis and skewness are also considered, along with energy spectral density based on Short-Time Fourier Transform (STFT) as a dynamic feature. All features are combined to form a high-dimensional vector, for example, [500, 200, 350, 60, kurtosis value, skewness value, STFT result...; 120, 40, 80, 25,...]. Vector concatenation is used to combine all features of the two signals into a higher-dimensional vector. Dimensionality reduction is then performed using PCA.

[0141] II. Health Index Mapping and Real-time Monitoring:

[0142] A dataset of 1000 sets containing contact pressure and arc energy characteristics and their corresponding expert-calibrated health indices from the past year was collected. This dataset was divided into a training set (70%), a validation set (15%), and a test set (15%). A random forest algorithm was used for training, and hyperparameters such as the number and depth of trees were optimized through cross-validation, ultimately achieving a mean squared error of 0.3 on the test set. The analytic hierarchy process (AHP) was used to determine the weights of each feature; for example, the weights of the three principal components were 0.4, 0.3, and 0.3, respectively.

[0143] The ARIMA model is used to model the historical trends of health indices, with safety upper and lower bounds set at 3 and 7, respectively. The health index is predicted for each day of the coming week; if the predicted value is below 3 or above 7 on any given day, an alarm is triggered and sent to the monitoring center.

[0144] III. Remaining life prediction:

[0145] Physical damage quantity model and data-driven model construction: Based on Holm's law, the cumulative damage quantity is calculated. A self-attention mechanism neural network is used to analyze the time-series relationship between comprehensive features and health index, outputting a nonlinear degradation term. Model parameters are adjusted based on historical data to better capture degradation trends. The drift and diffusion coefficients of the physical model and the data-driven model are combined, and their respective weights are determined to be 0.4 and 0.6 based on noise levels. Evolutionary game theory is applied to determine the dynamic degradation threshold, and the initial threshold is adjusted based on consensus to ensure accurate predictions while balancing the interests of all parties. Finally, based on the comprehensive degradation trajectory, the switch component is predicted to reach failure within the next 18 months, i.e., the remaining lifespan is approximately 18 months.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring the status of a switching component, characterized in that, include: S1. Extract single-signal features from the electrical parameters of the switching components of the power distribution control equipment, and perform feature fusion and dimensionality reduction to obtain the comprehensive features of the switching components; S2. Using the random forest algorithm, combined with historical data of the comprehensive characteristics of the switching components and expert calibration, a mapping from the comprehensive characteristics of the switching components to the health index is established; using a time series prediction model, the real-time trend of the health index is monitored, and an abnormal alarm is triggered when the preset conditions are reached. S3. Based on the comprehensive characteristics and health index of the switching components, and combined with the quantified value of the degradation trajectory and the dynamic degradation threshold of the switching components, the remaining life prediction results of the switching components are output, specifically including: Based on Holm's contact law, the cumulative damage to contact resistance caused by arc energy is calculated, and a physical damage model is constructed. The expression for the physical damage model is as follows: In the formula, Δ R ( t This represents the cumulative damage to the contact resistance caused by the electric arc energy. k Indicates the material coefficient. t Representing the time variable; using a self-attention neural network, the temporal relationship between the comprehensive characteristics of the switching components and the health index is analyzed, outputting a data-driven model of the switching components; the expression of the data-driven model is: In the formula, D 1( t ) represents a data-driven nonlinear degradation term; f This represents a neural network function based on a self-attention mechanism; X ( t This indicates the overall characteristics of the switching component. HI ( t () indicates the health index. t The time variable is represented; the adaptive weight is determined based on the noise level of the electrical parameters of the switching component; the drift coefficient and diffusion coefficient of the physical damage model and the data-driven model of the switching component are fused according to the adaptive weight to obtain the comprehensive degradation trajectory, which is used to represent the quantitative value of the current cumulative damage of the switching component. The comprehensive degradation trajectory includes: in the physical damage model of the switching component, the average growth rate of the cumulative damage value over time is used as the drift coefficient corresponding to the physical degradation term; by subtracting the actual measured degradation data of the switching component from the damage accumulation trend constructed by the physical damage model, several error terms are obtained, and the standard deviation of the error terms is calculated as the diffusion coefficient corresponding to the physical degradation term; in the data-driven model of the switching component, a self-attention mechanism neural network is used to extract the time-series dependence from the comprehensive features and health index of the switching component, outputting a data-driven nonlinear degradation term, and by analyzing the trend of the data-driven nonlinear degradation term, the drift coefficient corresponding to the data-driven is obtained; the self-attention mechanism neural network captures the inherent noise and nonlinear fluctuations in the comprehensive features and health index of the switching component, and generates the diffusion coefficient of the data-driven part; based on the physical degradation term and the drift coefficient and diffusion coefficient of the data-driven part, and combined with the weighted summation of adaptive weights, the fused drift coefficient and fused diffusion coefficient are obtained; The expression for the comprehensive degradation trajectory is: In the formula, Y ( T ) indicates in T The overall degradation status of the switching components at all times. Y (0) indicates the initial degradation state of the switching component; λ ( s ) represents the drift coefficient after fusion. σ ( s () represents the diffusion coefficient after fusion; dW ( s ) represents standard Brownian motion W ( s At that moment s The increment; The process involves identifying game participants, cost parameters, and configuring game strategies; constructing payoff matrices based on different strategies and cost parameters; calculating strategy proportion changes using dynamic evolution equations from evolutionary game theory and determining a preliminary dynamic degradation threshold based on Nash equilibrium points; statistically analyzing the urgency scores of different game participants based on historical maintenance records of the switch component; calculating the current consensus among all game participants; adjusting the preliminary dynamic degradation threshold based on the principle of amplifying the final game threshold when consensus is high and maintaining its stability when consensus is low, ultimately obtaining the dynamic degradation threshold; analyzing the comprehensive degradation trajectory of the switch component, and determining the moment when the quantified value of the current accumulated damage of the switch component in the comprehensive degradation trajectory equals the dynamic degradation threshold as the failure moment of the switch component, and obtaining the remaining lifespan of the switch component. The comprehensive characteristics of the switching components include the contact pressure and arc energy within the switching components.

2. The method for monitoring the status of a switching component according to claim 1, characterized in that, The process of extracting single-signal features from the electrical parameters of the switching components of the power distribution control equipment, and then performing feature fusion and dimensionality reduction to obtain the comprehensive features of the switching components includes: The contact pressure signal and arc energy signal of the switching component are acquired, and feature extraction and feature fusion are performed on the contact pressure signal and arc energy signal. Principal component analysis was used to reduce the dimensionality of the fused features, resulting in the comprehensive features of the switching components.

3. The method for monitoring the status of a switching component according to claim 1, characterized in that, The method of using the random forest algorithm, combined with historical data and expert calibration of the comprehensive characteristics of switching components, to establish a mapping from the comprehensive characteristics of switching components to the health index includes: By adding an expert-calibrated health index to the historical data of the comprehensive characteristics of the switching components, a dataset is obtained, which is then divided into a training set, a validation set, and a test set. Using the training set and based on the preset objective function, the random forest algorithm is used to fit the mapping from the comprehensive features of the switch components to the health index. The Analytic Hierarchy Process (AHP) is used to determine the weights of different features in the comprehensive features of the switching component, which will serve as the basis for feature selection in the Random Forest algorithm. During training, the optimal parameters of the random forest model are selected through cross-validation; the random forest model is validated and its performance is evaluated using the validation and test sets; the comprehensive features of the switching components acquired in real time are input into the random forest model, and the predicted health index is output.

4. The method for monitoring the status of a switching component according to claim 3, characterized in that, The use of the analytic hierarchy process (AHP) to determine the weights of different features in the comprehensive features of the switching component, as a basis for feature selection in the random forest algorithm, includes: The mapping from the comprehensive characteristics of switching components to the health index is defined as the target layer, the factors that affect the achievement of the target are defined as the criterion layer, and the characteristic variables are defined as the scheme layer. For each element in each layer, a judgment matrix is ​​constructed using a pairwise comparison method; the maximum eigenvalue and the corresponding eigenvector of each judgment matrix are calculated; the eigenvector represents the importance weight of each element relative to a certain element in the previous layer. The weights of each level are combined to obtain the total ranking weight of the scheme layer relative to the target layer; based on the total ranking weight, the features in the random forest algorithm are filtered.

5. The method for monitoring the status of a switching component according to claim 1, characterized in that, The method of using a time-series prediction model to monitor the real-time trend of health index changes and triggering anomaly alarms upon reaching preset conditions includes: A dynamic model of the health index is constructed using a time-series prediction model to capture the time-series variation patterns of the health index. The system uses a dynamic model of the health index to predict values ​​for future moments; it also pre-sets the upper and lower bounds of the health index for the switching components. If the health index of a switch component is greater than the upper limit or less than the lower limit at a certain future moment, it is predicted that the switch component will be in an abnormal state at that future moment, and the alarm module will be invoked to send an alarm signal to the monitoring center.

6. The method for monitoring the status of a switching component according to claim 1, characterized in that, The game participants include maintenance decision-makers, equipment users, operation and maintenance teams, and production teams; the game strategies include a preference for early maintenance and a preference for delayed maintenance; the cost parameters include the cost of early maintenance, the cost of over-maintenance, the cost of failure loss, and the cost of delayed maintenance risk; the payoff matrix includes conservative payoffs and aggressive payoffs.

7. The method for monitoring the status of a switching component according to claim 6, characterized in that, The urgency scores for different game participants, based on the historical maintenance records of the switching components, include: Collect historical maintenance records for switch components, including maintenance dates, maintenance types, and descriptions of problems found; Identify historical maintenance records directly relevant to each game participant and set urgency scoring criteria for different types of issues and maintenance activities; For each maintenance record, an urgency score is assigned to that maintenance record based on the urgency scoring criteria; For each game participant, the urgency scores corresponding to all maintenance records directly related to the game participant are summarized to obtain the game participant's urgency score.

8. A switching component status monitoring system, used to implement the switching component status monitoring method according to any one of claims 1-7, characterized in that, It includes a switch component feature acquisition module, a health index acquisition and anomaly warning module, and a lifespan prediction module; The switch component feature acquisition module is used to extract single signal features from the electrical parameters of the switch components of the power distribution control equipment, and perform feature fusion and dimensionality reduction to obtain the comprehensive features of the switch components. The health index acquisition and anomaly warning module is used to establish a mapping from the comprehensive characteristics of the switching components to the health index by using the random forest algorithm, combined with historical data and expert calibration of the comprehensive characteristics of the switching components; it uses a time series prediction model to monitor the real-time trend of the health index and triggers an anomaly alarm when the preset conditions are reached. The lifespan prediction module is used to output the remaining lifespan prediction results of the switching components based on the comprehensive characteristics and health index of the switching components, combined with the quantified value of the degradation trajectory of the switching components and the dynamic degradation threshold.

Images