Vacuum circuit breaker life state evaluation method and system based on data fusion

Abstract

This invention relates to the field of high-voltage electrical appliance life assessment technology, specifically a method and system for assessing the life status of vacuum circuit breakers based on data fusion. The method includes: acquiring historical operating conditions, online monitoring time series, and group failure data of the same model of vacuum circuit breaker; forming a multi-source fusion raw dataset through timestamp alignment and data cleaning; mining high-dimensional features using a deep feature extraction network to generate a comprehensive feature vector set reflecting mechanical, electrical, and insulation characteristics; constructing a multi-physics coupled aging damage model; obtaining preliminary life assessment results using the comprehensive feature vector set as input; fusing the preliminary results with prior life distribution based on an evidence theory fusion framework to generate a life status confidence distribution; classifying health status levels; calculating the expected value and confidence interval of remaining service life; and generating a structured assessment report. This solution improves the completeness and accuracy of the life status characterization of vacuum circuit breakers through the fusion of multi-source data and multiple results.

Description

Technical Field

[0001] This invention relates to the field of high-voltage electrical appliance life assessment technology, and in particular to a method and system for assessing the life status of vacuum circuit breakers based on data fusion. Background Technology

[0002] Current life condition assessments of vacuum circuit breakers often rely on a single data source, depending solely on historical operating data or online monitoring time-series data for status evaluation. Data processing often lacks timestamp alignment and data cleaning, feature extraction depends on superficial statistical analysis or manual experience screening, and fails to incorporate data on group failures of similar equipment. Furthermore, existing aging damage modeling for vacuum circuit breakers analyzes only a single physical field, neglecting the impact of the coupling effects of mechanical, electrical, and insulation multi-physical fields on the aging process. Life condition assessments rely solely on model calculations or single statistical distributions, failing to integrate multiple types of assessment information through an evidence-based theoretical fusion framework.

[0003] Existing assessment methods cannot standardize multi-source heterogeneous data, struggle to uncover hidden high-dimensional correlations within the data, and fail to comprehensively characterize the mechanical, electrical, and insulation properties of vacuum circuit breakers. Single-physics aging models cannot accurately reflect the cumulative damage and damage evolution trends of the equipment; single-form assessment results cannot reflect the confidence level of the lifespan status, accurately classify health status levels, quantify the expected value and confidence interval of the remaining effective service life, or generate standardized structured lifespan assessment reports. This invention addresses the shortcomings of existing technologies by achieving multi-source data fusion processing and high-dimensional feature mining, constructing a multi-physics coupled aging damage model, and using evidence theory to fuse assessment information, thereby improving the lifespan status assessment system for vacuum circuit breakers. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies by proposing a data fusion-based method and system for assessing the life condition of vacuum circuit breakers. To achieve the above objective, this invention adopts the following technical solution: a data fusion-based method for assessing the life condition of vacuum circuit breakers, comprising: Historical operating condition data, online monitoring time series data, and group failure data of the same type of equipment of vacuum circuit breaker are obtained. The historical operating condition data, online monitoring time series data, and group failure data are timestamped and cleaned to form a multi-source fused original dataset. High-dimensional feature mining is performed on the multi-source fusion original dataset based on a deep feature extraction network to generate a comprehensive feature vector set that reflects the mechanical, electrical and insulation characteristics of the vacuum circuit breaker. An aging damage model for a vacuum circuit breaker considering multi-physics coupling is constructed. The comprehensive feature vector set is used as the model input, and the cumulative damage and damage evolution trend of the vacuum circuit breaker are iteratively calculated to generate preliminary life status assessment results. Based on the evidence theory fusion framework, the preliminary life status assessment results generated by the aging damage model and the prior life distribution obtained from the statistics of mass failure data are fused to generate the fused life status confidence distribution of the vacuum circuit breaker. Based on the fused confidence distribution of the vacuum circuit breaker's life status, the health status level of the vacuum circuit breaker's life is classified, and the expected value and confidence interval of its remaining effective service life are calculated to generate a structured life assessment report.

[0005] As a further aspect of the present invention, the step of performing high-dimensional feature mining on the multi-source fusion original dataset based on a deep feature extraction network to generate a comprehensive feature vector set reflecting the mechanical, electrical, and insulation characteristics of the vacuum circuit breaker specifically includes: For the online monitoring time-series data in the multi-source fusion original dataset, the continuous wavelet transform method is used to extract multi-scale features in the time and frequency domain for the current curves of the opening and closing coils, the contact stroke curves, and the breaking current waveforms of the vacuum interrupter, respectively, generating a subset of mechanical vibration features, a subset of electromagnetic transient features, and a subset of arc characteristic features; for the historical operating condition data in the multi-source fusion original dataset, the cumulative effective value of breaking current, operating frequency, load rate, and ambient temperature and humidity sequence are extracted to generate a subset of operating statistical features; Using a stacked autoencoder network, unsupervised deep feature learning is performed on the mechanical vibration feature subset, electromagnetic transient feature subset, arc characteristic feature subset, and operational statistics feature subset, respectively, to obtain low-dimensional deep feature representations corresponding to each subset; The low-dimensional deep feature representations corresponding to each subset are concatenated, and a fully connected network layer is introduced to fuse and reduce the dimensionality of the concatenated features, ultimately outputting a comprehensive feature vector set.

[0006] As a further aspect of the present invention, the construction of the aging damage model of the vacuum circuit breaker considering multi-physics coupling, using the comprehensive feature vector set as model input, iteratively calculating the cumulative damage and damage evolution trend of the vacuum circuit breaker, and generating preliminary life status assessment results, specifically includes: A multi-physics coupling equation is established to describe the electrical wear of contacts, mechanical wear of mechanisms, and deterioration of insulation materials in vacuum circuit breakers. The features in the comprehensive feature vector set are mapped to the key input parameters in the coupling equation. Using the cumulative number of operations or equivalent breaking current of the vacuum circuit breaker as independent variables, the comprehensive feature vector set is input into the multi-physics coupling equation according to the operation time sequence, and the damage increment of each operation on contact mass loss, spring stiffness degradation and electrical trace depth of insulation is solved by numerical integration method. The damage increments are summed to obtain the cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation trace depth of the vacuum circuit breaker up to the current moment. The cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation trace depth are compared with preset failure thresholds to calculate the remaining life ratio of each key component at the current damage level. The remaining lifespan proportions of each key component are weighted and aggregated, with the weight determined by the importance of each component in the failure modes, to obtain the overall health index of the vacuum circuit breaker. The overall health index and its trend over time constitute the preliminary life status assessment results.

[0007] As a further aspect of the present invention, the information fusion based on the evidence theory fusion framework, which integrates the preliminary lifetime status assessment results generated by the aging damage model with the prior lifetime distribution obtained from the statistical analysis of mass failure data, specifically includes: From the collective failure data, the life distribution of the same type of vacuum circuit breaker under different operating conditions was statistically analyzed, and a Weibull distribution prior probability model of the equipment life was fitted. The overall health index in the preliminary life status assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current state of the equipment, which serves as evidence of the current state. A framework for identifying the life state of a vacuum circuit breaker is defined, the framework comprising multiple life state level ranges; Using the Weibull distribution prior probability model as the prior evidence body and the current state evidence body as the field evidence body, the confidence assignments of the field evidence body to each lifetime state level interval under the identification framework are calculated respectively. Using the synthesis rules of evidence theory, the confidence assignments given by the prior evidence and the on-site evidence are combined to obtain the fused joint confidence assignments for each lifetime state level interval, that is, the fused confidence distribution of the vacuum circuit breaker lifetime state.

[0008] As a further aspect of the present invention, the overall health index in the preliminary lifespan status assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current state of the equipment, specifically including: A nonlinear mapping model between the overall health index and the remaining effective service life of a vacuum circuit breaker is established. This nonlinear mapping model is trained using the full life cycle data of historical equipment of the same type. The overall health index of the vacuum circuit breaker at the current moment is extracted from the preliminary life status assessment results and input into the trained nonlinear mapping relationship model to obtain a single predicted value of the remaining life. Analyze the distribution pattern of historical prediction errors and construct an error distribution model centered on the single predicted value; The error distribution model is fitted and approximated using a B-spline function to generate a continuous probability density function with remaining lifetime as a random variable. This probability density function is the B-spline probability distribution model of the current state evidence body.

[0009] As a further aspect of the present invention, based on the fused confidence distribution of the vacuum circuit breaker's lifespan status, the health status levels of the vacuum circuit breaker's lifespan are classified, and the expected value and confidence interval of its remaining effective service life are calculated, specifically including: According to the operation and maintenance procedures for vacuum circuit breakers, a lifespan status level threshold is set to distinguish multiple states, including healthy, alert, abnormal, and critical. Based on the fused lifetime state confidence distribution of the vacuum circuit breaker, the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to each lifetime state level is calculated. The level with the highest cumulative confidence probability is determined as the current health status level of the vacuum circuit breaker. On the fused vacuum circuit breaker lifetime status confidence distribution, the mathematical expectation from the current time to the end of the lifetime is calculated, and the mathematical expectation value is the expected value of the remaining effective service life. Based on the fused confidence distribution of the vacuum circuit breaker's life status, the lower and upper confidence limits of the remaining effective service life are calculated using a set confidence level, thus forming its confidence interval. Based on the fused lifetime state confidence distribution of the vacuum circuit breaker, the cumulative confidence probability of the current state of the vacuum circuit breaker belonging to each lifetime state level is calculated, specifically including: The fused confidence distribution of the vacuum circuit breaker's lifetime status is regarded as the probability density function of the remaining lifetime random variable; The probability density function is integrated over the lifetime interval corresponding to each lifetime state level threshold, with the integration interval ranging from the minimum lifetime threshold to the maximum lifetime threshold of the lifetime state level. The result of the integral operation is the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to the lifetime state level. By iterating through all preset lifetime state levels and repeating the integration process, a set of cumulative confidence probability values ​​corresponding one-to-one with each lifetime state level is obtained.

[0010] As a further aspect of the present invention, the method of using a stacked autoencoder network to perform unsupervised deep feature learning on the mechanical vibration feature subset, electromagnetic transient feature subset, arc characteristic feature subset, and operational statistical feature subset respectively, to obtain low-dimensional deep feature representations corresponding to each subset, specifically including: For a subset of mechanical vibration features, a multi-layer autoencoder is constructed, whose input layer dimension is the same as that of the mechanical vibration feature subset. The high-dimensional input is compressed into a low-dimensional bottleneck layer through the nonlinear transformation of the encoder. The output of the bottleneck layer is the low-dimensional deep feature representation corresponding to the mechanical vibration feature subset. For the electromagnetic transient feature subset, a convolutional autoencoder is constructed. Its convolutional layer is used to extract the local correlation of features in space or time, the pooling layer is used for downsampling, and finally the low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset is obtained in the bottleneck layer. For the subset of arc characteristic features, a recurrent autoencoder is constructed, which contains long short-term memory network units to capture long-term dependencies in temporal features. Finally, the low-dimensional deep feature representation corresponding to the subset of arc characteristic features is obtained in the hidden state of the last time step. For a subset of operational statistical features, a standard multilayer perceptron autoencoder is constructed. The numerical features in the subset of operational statistical features are input, and their low-dimensional deep feature representation is obtained in the bottleneck layer.

[0011] As a further aspect of the present invention, the method of using the cumulative number of operations or equivalent breaking current of the vacuum circuit breaker as independent variables, inputting the comprehensive feature vector set into the multi-physics coupling equation according to the operation time sequence, and solving for the damage increment of each operation on contact mass loss, spring stiffness degradation, and insulation trace depth through numerical integration, specifically includes: From the comprehensive feature vector set, the feature vector corresponding to each operation event is parsed. The feature vector contains the breaking current value, arcing time, opening and closing speed and ambient temperature information for each operation. In the multiphysics coupling equation, the breaking current value and the arcing time are input into the electro-abrasion sub-model to calculate the incremental contact mass loss caused by the operation; In the multiphysics coupling equation, the opening and closing speed and the number of operations are input into the mechanical wear sub-model to calculate the equivalent spring stiffness degradation increment of the mechanism components caused by the operation; In the multiphysics coupling equation, the breaking current value, arcing time and ambient temperature are input into the insulation degradation sub-model to calculate the increment of the surface trace depth of the insulating material caused by the operation. Using the accumulation method in numerical integration, the increments of contact mass loss, spring stiffness degradation, and electrical trace depth of insulation obtained from each operation are accumulated to the corresponding cumulative damage.

[0012] As a further aspect of the present invention, the step of analyzing the distribution pattern of historical prediction errors and constructing an error distribution model centered on the single predicted value specifically includes: Collect historical data on the actual remaining life of vacuum circuit breakers of the same model under multiple different health indices, and the corresponding predicted values ​​obtained through the nonlinear mapping relationship model; Calculate the prediction error for each data point, where the prediction error is the difference between the actual remaining lifetime and the predicted remaining lifetime. Statistical analysis is performed on the prediction error of all data points to fit its probability distribution, which usually conforms to a normal distribution or a t-distribution with a mean of zero. The mean of the fitted error probability distribution is set to zero, and the variance is set to the sample variance of the historical prediction error, thereby constructing an error distribution model centered on the single prediction value and whose distribution shape is determined by the historical error.

[0013] As a further aspect of the present invention, the present invention also includes a vacuum circuit breaker life condition assessment system based on data fusion, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the vacuum circuit breaker life condition assessment method based on data fusion as described above.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By performing timestamp alignment and data cleaning on historical operating condition data, online monitoring time-series data, and group failure data of the same model of vacuum circuit breaker, temporal deviations and abnormal noise interference between multi-source data can be eliminated, forming a unified and adapted multi-source fused original dataset. High-dimensional feature mining is then performed on the multi-source fused original dataset using a deep feature extraction network, capturing deep correlation information that conventional feature extraction methods cannot obtain. The generated comprehensive feature vector set can fully cover the characterization information of the mechanical, electrical, and insulation characteristics of vacuum circuit breakers, avoiding the problem of one-sided state representation caused by a single data source.

[0015] An aging damage model for vacuum circuit breakers, considering multi-physics coupling, is constructed. Using a comprehensive feature vector set as input, iterative calculations of cumulative damage and damage evolution trends are performed, enabling preliminary lifespan assessment that closely aligns with the actual aging mechanism of vacuum circuit breakers. By fusing the preliminary assessment results from the aging damage model with prior lifespan distribution information obtained from group failure data statistics using an evidence-theory fusion framework, a quantitative lifespan confidence distribution is generated. Based on this confidence distribution, health status levels are classified, accurately matching the actual service status of the equipment. The calculated expected remaining effective service life and confidence intervals visually represent lifespan parameters, and the generated structured lifespan assessment report presents the assessment results and core parameters in a standardized manner. Attached Figure Description

[0016] Figure 1 This is a flowchart of the data fusion-based vacuum circuit breaker life status assessment method described in this invention; Figure 2 A flowchart for generating a comprehensive feature vector set based on a deep feature extraction network; Figure 3 This is a flowchart for information fusion based on the evidence theory fusion framework. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] See Figure 1 This invention provides a method for assessing the life status of vacuum circuit breakers based on data fusion, the specific method including: Historical operating condition data, online monitoring time-series data, and group failure data of the same model of vacuum circuit breaker are acquired. The acquired historical operating condition data, online monitoring time-series data, and group failure data are timestamped and cleaned to remove outliers and fill in missing values, thus forming a multi-source fused original dataset. High-dimensional feature mining is then performed on the multi-source fused original dataset using a deep feature extraction network. This network automatically learns and extracts deep-level features reflecting the mechanical, electrical, and insulation characteristics of the vacuum circuit breaker from the original data, generating a comprehensive feature vector set.

[0020] This deep feature extraction network is a modular and hierarchical system that follows the divide-and-conquer principle while integrating and unifying. Its workflow first uses signal processing and statistical methods to generate four specialized feature subsets from preprocessed multi-source data: mechanical vibration, electromagnetic transients, arc characteristics, and operational statistics. For each subset with different characteristics, four specially optimized stacked autoencoders are invoked in parallel for unsupervised deep feature learning and dimensionality reduction. A multi-layer autoencoder learns the principal component patterns of the vibration signal, a convolutional autoencoder extracts the local spatiotemporal patterns of the electromagnetic transient signal, a recurrent autoencoder captures the long-term dynamic evolution of arc characteristics, and a perceptron autoencoder abstracts the nonlinear correlations between operational statistics. The learned deep feature representations of each subset are concatenated, and a fully connected layer enables cross-physical domain feature interaction, weighted combination, and final dimensionality reduction, resulting in a comprehensive feature vector set that fully and deeply characterizes the equipment state and provides high-quality input for subsequent accurate physical modeling and life assessment.

[0021] A multi-physics coupling aging damage model for vacuum circuit breakers is constructed, which describes the equipment degradation process under the interaction of multiple physical fields such as electricity, magnetism, force, and heat. The comprehensive feature vector set is used as input to the model, and the cumulative damage and damage evolution trend of the vacuum circuit breaker are obtained through iterative calculation, generating a preliminary life status assessment result. Based on an evidence-theory fusion framework, the preliminary life status assessment result generated by the aging damage model and the prior life distribution obtained from collective failure data statistics are fused. By synthesizing evidence from different sources, a fused vacuum circuit breaker life status confidence distribution is generated, which integrates physical model knowledge and historical statistical knowledge. Based on the fused vacuum circuit breaker life status confidence distribution, the health status level of the vacuum circuit breaker life is classified, and the expected value and confidence interval of its remaining effective service life are calculated, generating a structured life assessment report to guide predictive maintenance of the equipment.

[0022] In one embodiment of the present invention, see [reference] Figure 2For the online monitoring time-series data in the multi-source fusion original dataset, continuous wavelet transform is used to extract multi-scale features in the time and frequency domain for the current curves of the opening and closing coils, the contact stroke curves, and the breaking current waveforms of the vacuum interrupter, generating mechanical vibration feature subsets, electromagnetic transient feature subsets, and arc characteristic feature subsets. For the historical operating condition data in the multi-source fusion original dataset, the cumulative effective value of breaking current, operating frequency, load rate, and ambient temperature and humidity sequences are extracted to generate operating statistical feature subsets. Using a stacked autoencoder network, unsupervised deep feature learning is performed on the mechanical vibration feature subset, electromagnetic transient feature subset, arc characteristic feature subset, and operating statistical feature subset, respectively. For the mechanical vibration feature subset, a multi-layer autoencoder is constructed, whose input layer dimension is the same as that of the mechanical vibration feature subset. The high-dimensional input is compressed into a low-dimensional bottleneck layer through the nonlinear transformation of the encoder, and the output of the bottleneck layer is the low-dimensional deep feature representation corresponding to the mechanical vibration feature subset. For the electromagnetic transient feature subset, a convolutional autoencoder is constructed. Its convolutional layers extract local spatial or temporal correlations of features, pooling layers perform downsampling, and finally, a low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset is obtained at the bottleneck layer. For the arc characteristic feature subset, a recurrent autoencoder is constructed, which includes long short-term memory network units to capture long-term dependencies in temporal features. Finally, a low-dimensional deep feature representation corresponding to the arc characteristic feature subset is obtained in the hidden state of the last time step. For the runtime statistical feature subset, a standard multilayer perceptron autoencoder is constructed. The numerical features from the runtime statistical feature subset are input, and their low-dimensional deep feature representations are obtained at the bottleneck layer. The low-dimensional deep feature representations extracted from each subset are concatenated along the feature dimension. The concatenation method involves directly linking the feature vectors corresponding to different subsets end-to-end along the channel dimension or feature dimension to form a fused high-dimensional feature vector. Subsequently, a fully connected network layer is introduced to perform nonlinear feature fusion and dimensionality compression on the concatenated high-dimensional features. Through linear transformation and feature fitting of the fully connected layer, redundant information is eliminated and key features are extracted, finally outputting a unified comprehensive feature vector set.

[0023] In the specific implementation, the online monitoring time-series data in the multi-source fusion original dataset are processed. For the current curves of the opening and closing coils, the contact stroke curves, and the breaking current waveforms of the vacuum interrupter, a continuous wavelet transform method is used to extract multi-scale features in the time-frequency domain, generating subsets of mechanical vibration features, electromagnetic transient features, and arc characteristic features. Specifically, the Morlet wavelet is selected as the mother wavelet function for the continuous wavelet transform. Convolution operations are performed on each monitoring curve under different scale parameters to generate a two-dimensional time-frequency graph. Statistical features are extracted from the time-frequency graph, including the mean, variance, energy, and entropy of the coefficients at each scale, forming corresponding feature subsets. The historical operating condition data in the multi-source fusion original dataset is processed to extract the cumulative effective value of breaking current, operation frequency, load rate and ambient temperature and humidity sequence, and generate an operating statistical feature subset. The cumulative effective value of breaking current is obtained by summing the square roots of the squares of each breaking current. The operation frequency is the number of opening and closing operations per unit time. The load rate is the ratio of average current to rated current. The ambient temperature and humidity sequence is the temperature and humidity measurement values ​​recorded in chronological order.

[0024] In some embodiments, a stacked autoencoder network is used to perform unsupervised deep feature learning on a subset of mechanical vibration features. A multi-layer autoencoder is constructed for the mechanical vibration feature subset, with its input layer dimension matching that of the feature subset. The high-dimensional input is compressed into a low-dimensional bottleneck layer through a nonlinear transformation of the encoder. The output of the bottleneck layer is the low-dimensional deep feature representation corresponding to the mechanical vibration feature subset. In some embodiments, the encoder consists of multiple fully connected layers, each followed by a nonlinear activation function. The decoder structure is symmetrical to this layer, and the loss function is the reconstruction error. The network is trained by minimizing the loss function using a backpropagation algorithm. After training, the decoder is discarded, and only the encoder is used to map the mechanical vibration feature subset into a low-dimensional deep feature representation.

[0025] In practice, a stacked autoencoder network is used to perform unsupervised deep feature learning on a subset of electromagnetic transient features. A convolutional autoencoder is constructed for this subset, using its convolutional layers to extract local spatial or temporal correlations of features, and pooling layers to perform downsampling. Finally, a low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset is obtained at the bottleneck layer. The encoder part of the convolutional autoencoder includes convolutional layers and pooling layers. The convolutional layers use one-dimensional convolutional kernels to slide across the feature sequence to extract local patterns, and the pooling layers use max pooling to reduce the feature dimensionality. The decoder part consists of deconvolutional layers and upsampling layers for reconstructing the input. After network training, the feature vector of the bottleneck layer is the low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset.

[0026] Optionally, unsupervised deep feature learning is performed on the arc characteristic feature subset using a stacked autoencoder network. For the arc characteristic feature subset, a recurrent autoencoder is constructed, containing long short-term memory (LSM) network units to capture long-term dependencies in the temporal features. Finally, the low-dimensional deep feature representation corresponding to the arc characteristic feature subset is obtained from the hidden state of the last time step. The encoder of the recurrent autoencoder is a multi-layer LSM network that processes the input arc characteristic feature subset sequence step by step. The hidden state of the last time step is used as the context encoding for the entire sequence. The decoder is another LSM network that attempts to reconstruct the input sequence using the context encoding as the initial state. After training, the hidden state of the encoder at the last time step is the low-dimensional deep feature representation corresponding to the arc characteristic feature subset.

[0027] In practice, a stacked autoencoder network is used to perform unsupervised deep feature learning on a subset of running statistical features. For this subset, a standard multilayer perceptron autoencoder is constructed. The numerical features from the subset are input, and a low-dimensional deep feature representation is obtained at the bottleneck layer. The multilayer perceptron autoencoder consists of an input layer, an encoder with multiple hidden layers, a bottleneck layer, and a symmetrical decoder. Its objective function is to minimize the mean squared error between the input features and the reconstructed features. After training convergence, the activation values ​​of the bottleneck layer constitute the low-dimensional deep feature representation corresponding to the subset of running statistical features.

[0028] It is understandable that after obtaining the low-dimensional deep feature representations corresponding to each feature subset, feature fusion is required. This involves concatenating the low-dimensional deep feature representations corresponding to the mechanical vibration feature subset, the electromagnetic transient feature subset, the arc characteristic feature subset, and the operational statistics feature subset. The concatenation operation links the four independent feature vectors end-to-end, combining them into a higher-dimensional joint feature vector. Subsequently, a fully connected network layer is introduced to fuse and reduce the dimensionality of the concatenated joint feature vector. The fully connected network layer receives the concatenated vector as input and, through linear transformation and nonlinear activation functions, maps the high-dimensional features to a lower-dimensional vector space. The transformation relationship can be expressed as: in: This represents the concatenated joint feature vector. and These are the weight matrix and bias vector of the fully connected layer, respectively. It is a non-linear activation function. It is the output vector of the fully connected network layer. It can be understood that the output vector of the fully connected network layer is the final generated comprehensive feature vector set, which integrates deep abstract features from multiple aspects such as mechanics, electromagnetism, electric arcs and operation statistics.

[0029] In one embodiment of the present invention, a multiphysics coupling equation is established to describe the electrical wear of the contacts, mechanical wear of the mechanism, and degradation of the insulation material of a vacuum circuit breaker. The features in the comprehensive feature vector set are mapped to key input parameters in the coupling equation. Using the cumulative number of operations or the equivalent breaking current of the vacuum circuit breaker as independent variables, the comprehensive feature vector set is input into the multiphysics coupling equation according to the operation time sequence. From the comprehensive feature vector set, the feature vector corresponding to each operation event is analyzed. The feature vector contains the breaking current value, arcing time, opening and closing speed, and ambient temperature information for each operation. In the multiphysics coupling equation, the breaking current value and arcing time are input into the electrical wear sub-model to calculate the increment of contact mass loss caused by the operation. In the multiphysics coupling equation, the opening and closing speed and the number of operations are input into the mechanical wear sub-model to calculate the increment of equivalent spring stiffness degradation of the mechanism components caused by the operation. In the multiphysics coupling material equation, the breaking current value, arcing time, and ambient temperature are input into the insulation degradation sub-model to calculate the increment of surface electrical tracking depth of the insulation material caused by the operation. Using the accumulation method in numerical integration, the increments of contact mass loss, spring stiffness degradation, and insulation tracking depth calculated for each operation are added to the corresponding cumulative damage values. The numerical integration method is then used to solve for the damage increments of each operation on contact mass loss, spring stiffness degradation, and insulation tracking depth. These damage increments are then summed to obtain the cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation tracking depth of the vacuum circuit breaker up to the current time. These cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation tracking depth are compared with preset failure thresholds to calculate the remaining life ratio of each key component at the current damage level. The remaining life ratios of each key component are then weighted and aggregated, with the weights determined by the importance of each component in the failure mode, to obtain the overall health index of the vacuum circuit breaker. This overall health index and its trend over time constitute the preliminary life status assessment result.

[0030] In practical implementation, the aging damage model of the vacuum circuit breaker is a mathematical model based on physical mechanisms. Its core consists of multi-physics coupling equations describing contact electrical wear, mechanical wear, and insulation material degradation. Multi-physics coupling equations describing contact electrical wear, mechanical wear, and insulation material degradation of the vacuum circuit breaker are established, and the features in the comprehensive feature vector set are mapped to key input parameters in the coupling equations. The multi-physics coupling equations are composed of an electrical wear sub-model, a mechanical wear sub-model, and an insulation degradation sub-model coupled through shared physical variables. Key input parameters in the coupling equations include breaking current, arcing time, opening and closing speed, ambient temperature, and cumulative number of operations. The values ​​in the comprehensive feature vector set are analyzed and assigned to the corresponding key input parameters. Using the cumulative number of operations or the equivalent breaking current of the vacuum circuit breaker as independent variables, the comprehensive feature vector set is input into the multi-physics coupling equations according to the operation time sequence. In practice, the feature vector corresponding to each operation event is analyzed from the comprehensive feature vector set. The feature vector contains the breaking current value, arcing time, opening and closing speed and ambient temperature information for each operation. The analysis process is carried out according to the preset dimension order and physical meaning mapping table in the feature vector.

[0031] In some embodiments, the breaking current value and arcing time are input into the electro-wear sub-model in the multiphysics coupling equation to calculate the incremental contact mass loss caused by the operation. The electro-wear sub-model is based on the arc energy dissipation theory; the incremental contact mass loss caused by each operation is proportional to the product of the square of the breaking current value and the arcing time, with the proportionality coefficient determined by the contact material properties. In some embodiments, the opening and closing speeds and the number of operations are input into the mechanical wear sub-model in the multiphysics coupling equation to calculate the incremental equivalent spring stiffness degradation of the mechanism components caused by the operation. The mechanical wear sub-model is based on the fatigue cumulative damage theory, mapping the opening and closing speeds to the stress amplitude experienced by the mechanism components. Each operation is considered a stress cycle; the incremental equivalent spring stiffness degradation is proportional to the power of the stress amplitude and exhibits a non-linear increase with the increase of the cumulative number of operations.

[0032] In practical implementation, the breaking current, arcing time, and ambient temperature are input into the insulation degradation sub-model in the multi-physics coupled material equation to calculate the increment of the electrical trace depth on the insulating material surface caused by the operation. The insulation degradation sub-model comprehensively considers the electrothermal aging effect. The breaking current and arcing time jointly determine the thermal shock energy of the arc on the insulating material surface. Ambient temperature affects the initial aging state of the material. The increment of electrical trace depth has an exponential relationship with the thermal shock energy and is corrected by ambient temperature. Optionally, the accumulation method in numerical integration is used to accumulate the increment of contact mass loss, spring stiffness degradation, and electrical trace depth of the insulating component calculated for each operation to the corresponding cumulative damage amount. The numerical integration method is used to solve for the damage increment of contact mass loss, spring stiffness degradation, and electrical trace depth of the insulating component for each operation. The numerical integration process is sequential and iterative. The cumulative damage amount after the nth operation is equal to the cumulative damage amount after the (n-1)th operation plus the damage increment calculated for the nth operation.

[0033] It is understandable that the incremental damage is accumulated and summed to obtain the cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation tracking depth of the vacuum circuit breaker up to the current moment. This accumulation and summation operation is performed throughout the entire service history of the vacuum circuit breaker, starting from the first operation and updating the values ​​of the three cumulative damage quantities after each operation. In practice, the cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation tracking depth are compared with preset failure thresholds to calculate the remaining life ratio of each key component at the current damage level. The failure threshold is a technical indicator determined in advance through experiments or simulations. When the cumulative damage reaches the failure threshold, the component is considered to have failed functionally. The formula for calculating the remaining life ratio is: in: This represents the remaining lifespan percentage of the i-th component. This represents the current cumulative damage amount of the i-th component. This represents the preset failure threshold corresponding to the i-th component, where the subscript i represents the contact mass loss, spring stiffness degradation, and electrical trace depth of the insulation component, respectively.

[0034] Optionally, the remaining lifespan proportions of various key components are weighted and aggregated, with the weights determined by the importance of each component in the failure modes, to obtain the overall health index of the vacuum circuit breaker. The weighting coefficients are obtained through failure mode and effects analysis (FMEA) of historical failure cases, and the overall health index is calculated as the weighted arithmetic mean of the remaining lifespan proportions of each component. It can be understood that the overall health index and its trend over time constitute the preliminary lifespan status assessment result. The overall health index is a scalar value between 0 and 1; the closer its value is to 1, the better the health status of the vacuum circuit breaker. The trend over time is formed by recording the overall health index values ​​after each operation.

[0035] In one embodiment of the present invention, see [reference] Figure 3 From the collective failure data, the lifespan distribution of vacuum circuit breakers of the same model under different operating conditions is statistically analyzed, and a Weibull distribution prior probability model of the equipment lifespan is fitted. The overall health index in the preliminary lifespan status assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current equipment status, serving as the current status evidence. A framework for identifying the lifespan status of vacuum circuit breakers is defined, which includes multiple lifespan status level intervals. Using the Weibull distribution prior probability model as the prior evidence and the current status evidence as the field evidence, the confidence assignments for each lifespan status level interval under the identification framework are calculated for both the prior and field evidence. Using the synthesis rules of evidence theory, the confidence assignments given by the prior evidence and the field evidence are combined to obtain the fused joint confidence assignments for each lifespan status level interval, i.e., the fused vacuum circuit breaker lifespan status confidence distribution.

[0036] In practical implementation, the lifespan distribution of the same type of vacuum circuit breaker under different operating conditions is statistically analyzed from the mass failure data. A Weibull distribution prior probability model of the equipment lifespan is fitted. The mass failure data comes from the operation records of multiple scrapped vacuum circuit breakers of the same type, recording the final failure time or equivalent number of operations for each device. By fitting the Weibull distribution to these lifespan data, the shape parameter β and scale parameter η describing the statistical law of the mass lifespan of this type of vacuum circuit breaker are obtained. The overall health index in the preliminary lifespan status assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current state of the equipment, serving as evidence of the current state. The mapping process is completed using a pre-trained Gaussian process regression model. This model takes the overall health index as input and outputs the mean and variance of the predicted remaining lifespan. Then, a normal distribution with the mean as the expectation and the variance representing the uncertainty is constructed, and the probability density of this normal distribution is fitted using a cubic B-spline function. Define a framework for identifying the life status of vacuum circuit breakers. The framework includes multiple life status level ranges, which are divided according to industry standards or operation and maintenance experience. For example, they are divided into four levels: "Sufficient Life", "Life Attention", "Life Warning", and "Life Critical", each corresponding to a different range of remaining life values.

[0037] In some embodiments, the Weibull distribution prior probability model is used as the prior evidence body, and the current state evidence body is used as the field evidence body. Confidence assignments for each lifetime state level interval are calculated for both the prior evidence body and the field evidence body within the identification framework. For the prior evidence body, its confidence assignment is obtained by calculating the probability mass of the Weibull distribution over each lifetime state level interval, i.e., integrating the probability density function of the Weibull distribution over the corresponding remaining lifetime interval. In some embodiments, for the field evidence body, its confidence assignment is obtained by calculating the integral of the B-spline probability distribution model over each lifetime state level interval. The basic confidence assignments for the propositions in the identification framework for both types of evidence bodies are shown in Table 1. Table 1: Basic Confidence Allocation of Evidence to Lifespan Status Levels In practical implementation, the composition rules of evidence theory are used to combine the confidence assignments given by prior evidence and on-site evidence. Dempster's composition rule is the core of information fusion, and its mathematical expression is: in: It represents any proposition or combination of propositions in the identification framework (i.e., the lifespan state level range). This indicates the effect of fusion on the proposition. Joint confidence allocation, and These respectively represent the verification of the evidence body and the on-site evidence body regarding the proposition. and Basic confidence assignment, symbol This indicates that the proposition satisfies With proposition The intersection equals the proposition Summing all pairs of propositions, This is the normalization coefficient, also often called the conflict coefficient, and its calculation formula is: It is used to measure the magnitude of conflict between two pieces of evidence.

[0038] In one embodiment of the present invention, a nonlinear mapping model between the overall health index of a vacuum circuit breaker and its remaining effective service life is established. This nonlinear mapping model is trained using full lifecycle data of historical similar equipment. The overall health index of the vacuum circuit breaker at the current moment is extracted from the preliminary lifespan assessment results and input into the trained nonlinear mapping model to obtain a single predicted value of the remaining service life. The distribution pattern of historical prediction errors is analyzed, and an error distribution model centered on this single predicted value is constructed. Actual remaining service life data of historical similar vacuum circuit breakers under multiple different health indices are collected, along with the corresponding predicted values ​​obtained through the nonlinear mapping model. The prediction error corresponding to each data point is calculated; this prediction error is the difference between the actual remaining service life and the predicted remaining service life. Statistical analysis is performed on the prediction errors of all data points, and their probability distribution is fitted. The probability distribution typically conforms to a normal distribution with a mean of zero or a t-distribution. The mean of the fitted error probability distribution is set to zero, and the variance is set to the sample variance of the historical prediction errors, thereby constructing an error distribution model centered on the single predicted value, with its distribution shape determined by historical errors. The error distribution model is fitted and approximated using a B-spline function to generate a continuous probability density function with remaining lifetime as a random variable. This probability density function is the B-spline probability distribution model of the current state evidence body.

[0039] In practical implementation, a nonlinear mapping model is established between the overall health index of the vacuum circuit breaker and its remaining effective service life. This model is trained using full lifecycle data of historical equipment of the same type. The training data includes records of the overall health index of multiple vacuum circuit breakers at different points in their lifecycles, along with the corresponding actual remaining service life from that point until equipment failure. The nonlinear mapping model can be a deep feedforward neural network. Its input layer receives the overall health index, passes through multiple hidden layers with nonlinear activation functions, and outputs a predicted value for the remaining service life. The network parameters are trained by minimizing the mean square error between the predicted and actual values. The overall health index of the vacuum circuit breaker at the current moment is extracted from the preliminary lifespan assessment results and input into the trained nonlinear mapping model to obtain a single predicted value for the remaining service life. For example, if the current overall health index of the vacuum circuit breaker is 0.75, inputting it into the trained neural network model will output a specific value, such as a predicted remaining service life of 8.3 years.

[0040] In some embodiments, the distribution pattern of historical prediction errors is analyzed, and an error distribution model centered on a single predicted value is constructed. Real remaining life data of the same model of vacuum circuit breaker under multiple different health indices are collected, along with the corresponding predicted values ​​obtained through a nonlinear mapping model, forming a dataset for error analysis. In some embodiments, an example of the dataset is shown in Table 2, which displays five sets of historical real values, model predicted values, and calculated errors: Table 2: Historical Prediction Error Analysis Data Table In practice, the prediction error for each data point is calculated. The prediction error is the difference between the actual remaining lifetime and the predicted remaining lifetime. The calculated error value is recorded in the "Prediction Error" column of Table 2. Statistical analysis is performed on the prediction errors of all data points to fit their probability distribution. The probability distribution typically conforms to a normal distribution or a t-distribution with a mean of zero. The fitting process involves calculating the sample mean and sample variance of the error and using maximum likelihood estimation or the method of moments to determine the specific parameters of the distribution. Essentially, by setting the mean of the fitted error probability distribution to zero and the variance to the sample variance of historical prediction errors, an error distribution model is constructed centered on a single predicted value, with the distribution shape determined by historical errors. This error distribution model describes the possible fluctuation range of the actual remaining lifetime under a given predicted value.

[0041] Optionally, a B-spline function can be used to fit and approximate the error distribution model, generating a continuous probability density function with remaining lifetime as the random variable. (B-spline probability density function) It can be represented as a linear combination of a series of B-spline basis functions, i.e.: in: Represents the remaining lifespan random variable. It is the k-th B-spline basis function The corresponding coefficients, It is the degree of the B-spline. It represents the number of basis functions. This can be understood as the coefficients... This is obtained by solving a constrained optimization problem, which aims to minimize the difference between the B-spline function and the target error distribution model, and is subject to constraints. The integral over the entire domain is 1 to ensure that it is a valid probability density function. The resulting continuous probability density function with remaining lifespan as the random variable is the B-spline probability distribution model of the current state evidence body. This model expresses the uncertainty in predicting remaining lifespan based on the current health index as a smooth curve.

[0042] In one embodiment of the present invention, according to the operation and maintenance procedures for vacuum circuit breakers, a lifespan state level threshold is set to distinguish between multiple states such as healthy, alert, abnormal, and critical. Based on the fused lifespan state confidence distribution of the vacuum circuit breaker, the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to each lifespan state level is calculated.

[0043] The formula for calculating the cumulative confidence probability is: in: This represents the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to the i-th lifetime state level. The confidence distribution of the life state of the fused vacuum circuit breaker is represented and is considered as the probability density function of the remaining life random variable t in the calculation. and These represent the lower and upper limits of the preset lifespan range for the i-th lifespan state level, respectively. Integration operation. That is, in the probability density function Above, for variable t from arrive The definite integral yields the probability that the state belongs to that interval. By iterating through all preset lifetime state levels and calculating each level individually, a set of cumulative confidence probabilities corresponding to each level can be obtained. And satisfy .

[0044] The fused vacuum circuit breaker life status confidence distribution is considered as a probability density function of the remaining lifespan random variable. The probability density function is integrated over the lifespan interval corresponding to each lifespan status level threshold, with the integration interval ranging from the minimum to the maximum lifespan threshold of the lifespan status level. The result of the integration is the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to that lifespan status level. This integration process is repeated for all preset lifespan status levels to obtain a set of cumulative confidence probability values ​​corresponding one-to-one with each lifespan status level. The level with the highest cumulative confidence probability is determined as the current health status level of the vacuum circuit breaker. The expected value from the current time to the end of the lifespan is calculated on the fused vacuum circuit breaker life status confidence distribution; this expected value is the expected value of the remaining effective service life. Finally, the lower and upper confidence limits of the remaining effective service life are calculated on the fused vacuum circuit breaker life status confidence distribution at a set confidence level, forming its confidence interval.

[0045] In practical implementation, the operation and maintenance procedures for vacuum circuit breakers set thresholds to distinguish between multiple lifespan status levels, such as healthy, caution, abnormal, and critical. The procedures clearly define the remaining lifespan range corresponding to different health states. For example, a remaining lifespan greater than 15 years is considered "healthy," between 8 and 15 years is "caution," between 3 and 8 years is "abnormal," and less than 3 years is "critical." Based on the fused confidence distribution of vacuum circuit breaker lifespan status, the cumulative confidence probability of the current state of the vacuum circuit breaker belonging to each lifespan status level is calculated. The fused confidence distribution of vacuum circuit breaker lifespan status is considered as the probability density function of the remaining lifespan random variable. The probability density function is integrated over the lifespan interval corresponding to each lifespan status level threshold, from the minimum to the maximum lifespan threshold. For example, calculating the cumulative confidence probability for the "caution" status level involves definite integration of the probability density function over the remaining lifespan interval of 8 to 15 years. The result of the integration operation is the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to a certain lifespan state level. By iterating through all preset lifespan state levels and repeating the integration operation, a set of cumulative confidence probability values ​​corresponding one-to-one with each lifespan state level is obtained, and the sum of these probability values ​​is 1.

[0046] In some embodiments, the level with the highest cumulative confidence probability is determined as the current health status level of the vacuum circuit breaker. By comparing the cumulative confidence probability values ​​corresponding to each lifetime status level, the level corresponding to the maximum value is selected as the final determination result. In some embodiments, the expected value from the current time to the end of the lifetime is calculated on the fused vacuum circuit breaker lifetime status confidence distribution. The expected value is the expected value of the remaining effective service life. The expected value is calculated by integrating the product of the remaining lifetime variable and the probability density function over the interval from zero to positive infinity. The calculation formula is as follows: in: This represents the expected value of the remaining effective service life. Represents the remaining lifespan variable. The probability density function represents the confidence distribution of the life state of the integrated vacuum circuit breaker.

[0047] In practical implementation, the lower and upper confidence limits of the remaining effective service life are calculated based on the merged confidence distribution of the vacuum circuit breaker's life status, using a set confidence level, to form its confidence interval. The set confidence level is a pre-selected probability value, such as a 95% confidence level. Optionally, calculating the lower and upper confidence limits requires solving an integral equation such that the integral of the probability density function from the lower confidence limit to positive infinity equals... And the integral of the probability density function over zero to the upper confidence limit is equal to It is understandable that for asymmetric probability distributions, the quantile method is typically used to determine confidence intervals, finding two quantiles such that the area under the integral of the probability density function between these two points equals the set confidence level. Similarly, it is understandable that the final expected remaining effective service life, the confidence interval, and the determined health status level together constitute the core quantitative conclusions of the life assessment report.

[0048] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for assessing the life condition of vacuum circuit breakers based on data fusion, characterized in that, Includes the following steps: Historical operating condition data, online monitoring time series data, and group failure data of the same type of equipment of vacuum circuit breaker are obtained. The historical operating condition data, online monitoring time series data, and group failure data are timestamped and cleaned to form a multi-source fused original dataset. High-dimensional feature mining is performed on the multi-source fusion original dataset based on a deep feature extraction network to generate a comprehensive feature vector set that reflects the mechanical, electrical and insulation characteristics of the vacuum circuit breaker. An aging damage model for a vacuum circuit breaker considering multi-physics coupling is constructed. The comprehensive feature vector set is used as the model input. The cumulative damage and damage evolution trend of the vacuum circuit breaker are iteratively calculated to generate preliminary life status assessment results, specifically including: A multi-physics coupling equation is established to describe the electrical wear of contacts, mechanical wear of mechanisms, and deterioration of insulation materials in vacuum circuit breakers. The features in the comprehensive feature vector set are mapped to the key input parameters in the coupling equation. Using the cumulative number of operations or equivalent breaking current of the vacuum circuit breaker as independent variables, the comprehensive feature vector set is input into the multi-physics coupling equation according to the operation time sequence, and the damage increment of each operation on contact mass loss, spring stiffness degradation and electrical trace depth of insulation is solved by numerical integration method. The damage increments are summed to obtain the cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation trace depth of the vacuum circuit breaker up to the current moment. The cumulative contact mass loss, cumulative spring stiffness degradation, and cumulative insulation trace depth are compared with preset failure thresholds to calculate the remaining life ratio of each key component at the current damage level. The remaining life ratios of each key component are weighted and aggregated, where the weight is determined by the importance of each component in the failure mode, to obtain the overall health index of the vacuum circuit breaker. The overall health index and its trend over time constitute the preliminary life status assessment results. Based on the evidence theory fusion framework, the preliminary life status assessment results generated by the aging damage model and the prior life distribution obtained from the statistical analysis of mass failure data are fused to generate a fused confidence distribution of the vacuum circuit breaker's life status, specifically including: From the collective failure data, the life distribution of the same type of vacuum circuit breaker under different operating conditions was statistically analyzed, and a Weibull distribution prior probability model of the equipment life was fitted. The overall health index in the preliminary life status assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current state of the equipment, which serves as evidence of the current state. A framework for identifying the life state of a vacuum circuit breaker is defined, the framework comprising multiple life state level ranges; Using the Weibull distribution prior probability model as the prior evidence body and the current state evidence body as the field evidence body, the confidence assignments of the field evidence body to each lifetime state level interval under the identification framework are calculated respectively. Using the synthesis rules of evidence theory, the confidence assignments given by the prior evidence and the on-site evidence are combined to obtain the fused joint confidence assignments for each lifetime state level interval, that is, the fused vacuum circuit breaker lifetime state confidence distribution. Based on the fused confidence distribution of the vacuum circuit breaker's lifespan status, the health status levels of the vacuum circuit breaker's lifespan are classified, and the expected value and confidence interval of its remaining effective service life are calculated to generate a structured lifespan assessment report, specifically including: According to the operation and maintenance procedures for vacuum circuit breakers, a lifespan status level threshold is set to distinguish multiple states, including healthy, alert, abnormal, and critical. Based on the fused lifetime state confidence distribution of the vacuum circuit breaker, the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to each lifetime state level is calculated. The level with the highest cumulative confidence probability is determined as the current health status level of the vacuum circuit breaker. On the fused vacuum circuit breaker lifetime status confidence distribution, the mathematical expectation from the current time to the end of the lifetime is calculated, and the mathematical expectation value is the expected value of the remaining effective service life. Based on the fused confidence distribution of the vacuum circuit breaker's life status, the lower and upper confidence limits of the remaining effective service life are calculated using a set confidence level, thus forming its confidence interval. Based on the fused lifetime state confidence distribution of the vacuum circuit breaker, the cumulative confidence probability of the current state of the vacuum circuit breaker belonging to each lifetime state level is calculated, specifically including: The fused confidence distribution of the vacuum circuit breaker's lifetime status is regarded as the probability density function of the remaining lifetime random variable; The probability density function is integrated over the lifetime interval corresponding to each lifetime state level threshold, with the integration interval ranging from the minimum lifetime threshold to the maximum lifetime threshold of the lifetime state level. The result of the integral operation is the cumulative confidence probability that the current state of the vacuum circuit breaker belongs to the lifetime state level. By iterating through all preset lifetime state levels and repeating the integration process, a set of cumulative confidence probability values ​​corresponding one-to-one with each lifetime state level is obtained.

2. The method for assessing the life status of vacuum circuit breakers based on data fusion according to claim 1, characterized in that, The high-dimensional feature mining of the multi-source fusion original dataset based on the deep feature extraction network generates a comprehensive feature vector set reflecting the mechanical, electrical, and insulation characteristics of the vacuum circuit breaker, specifically including: For the online monitoring time-series data in the multi-source fusion original dataset, the continuous wavelet transform method is used to extract time-frequency domain multi-scale features for the opening and closing coil current curves, contact stroke curves, and vacuum interruptor current waveforms, respectively, to generate mechanical vibration feature subsets, electromagnetic transient feature subsets, and arc characteristic feature subsets. From the historical operating condition data in the multi-source fusion original dataset, extract the cumulative effective value of breaking current, operation frequency, load rate and ambient temperature and humidity sequence to generate a subset of operating statistical features; Using a stacked autoencoder network, unsupervised deep feature learning is performed on the mechanical vibration feature subset, electromagnetic transient feature subset, arc characteristic feature subset, and operational statistics feature subset, respectively, to obtain low-dimensional deep feature representations corresponding to each subset; The low-dimensional deep feature representations corresponding to each subset are concatenated, and a fully connected network layer is introduced to fuse and reduce the dimensionality of the concatenated features, ultimately outputting a comprehensive feature vector set.

3. The method for assessing the life status of a vacuum circuit breaker based on data fusion according to claim 2, characterized in that, The overall health index in the preliminary lifespan assessment results generated by the aging damage model is mapped to a B-spline probability distribution model of the remaining lifespan corresponding to the current state of the equipment, specifically including: A nonlinear mapping model between the overall health index and the remaining effective service life of a vacuum circuit breaker is established. This nonlinear mapping model is trained using the full life cycle data of historical equipment of the same type. The overall health index of the vacuum circuit breaker at the current moment is extracted from the preliminary life status assessment results and input into the trained nonlinear mapping relationship model to obtain a single predicted value of the remaining life. Analyze the distribution pattern of historical prediction errors and construct an error distribution model centered on the single predicted value; The error distribution model is fitted and approximated using a B-spline function to generate a continuous probability density function with remaining lifetime as a random variable. This probability density function is the B-spline probability distribution model of the current state evidence body.

4. The method for assessing the life status of a vacuum circuit breaker based on data fusion according to claim 3, characterized in that, The method utilizes a stacked autoencoder network to perform unsupervised deep feature learning on the mechanical vibration feature subset, electromagnetic transient feature subset, arc characteristic feature subset, and operational statistical feature subset, respectively, to obtain low-dimensional deep feature representations corresponding to each subset, specifically including: For a subset of mechanical vibration features, a multi-layer autoencoder is constructed, whose input layer dimension is the same as that of the mechanical vibration feature subset. The high-dimensional input is compressed into a low-dimensional bottleneck layer through the nonlinear transformation of the encoder. The output of the bottleneck layer is the low-dimensional deep feature representation corresponding to the mechanical vibration feature subset. For the electromagnetic transient feature subset, a convolutional autoencoder is constructed. Its convolutional layer is used to extract the local correlation of features in space or time, the pooling layer is used for downsampling, and finally the low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset is obtained in the bottleneck layer. For the subset of arc characteristic features, a recurrent autoencoder is constructed, which contains long short-term memory network units to capture long-term dependencies in temporal features. Finally, the low-dimensional deep feature representation corresponding to the subset of arc characteristic features is obtained in the hidden state of the last time step. For the electromagnetic transient feature subset, a convolutional autoencoder is constructed. Its convolutional layer is used to extract the local correlation of features in space or time, the pooling layer is used for downsampling, and finally the low-dimensional deep feature representation corresponding to the electromagnetic transient feature subset is obtained in the bottleneck layer. For the subset of arc characteristic features, a recurrent autoencoder is constructed, which contains long short-term memory network units to capture long-term dependencies in temporal features. Finally, the low-dimensional deep feature representation corresponding to the subset of arc characteristic features is obtained in the hidden state of the last time step. For a subset of operational statistical features, a standard multilayer perceptron autoencoder is constructed. The numerical features in the subset of operational statistical features are input, and their low-dimensional deep feature representation is obtained in the bottleneck layer.

5. The method for assessing the life status of a vacuum circuit breaker based on data fusion according to claim 4, characterized in that, The method uses the cumulative number of operations or equivalent breaking current of the vacuum circuit breaker as independent variables, inputs the comprehensive feature vector set into the multiphysics coupling equation according to the operation time sequence, and solves the damage increment of each operation on contact mass loss, spring stiffness degradation, and insulation trace depth through numerical integration. Specifically, this includes: From the comprehensive feature vector set, the feature vector corresponding to each operation event is parsed. The feature vector contains the breaking current value, arcing time, opening and closing speed and ambient temperature information for each operation. In the multiphysics coupling equation, the breaking current value and the arcing time are input into the electro-abrasion sub-model to calculate the incremental contact mass loss caused by the operation; In the multiphysics coupling equation, the opening and closing speed and the number of operations are input into the mechanical wear sub-model to calculate the equivalent spring stiffness degradation increment of the mechanism components caused by the operation; In the multiphysics coupling equation, the breaking current value, arcing time and ambient temperature are input into the insulation degradation sub-model to calculate the increment of the surface trace depth of the insulating material caused by the operation. Using the accumulation method in numerical integration, the increments of contact mass loss, spring stiffness degradation, and electrical trace depth of insulation obtained from each operation are accumulated to the corresponding cumulative damage.

6. The method for assessing the life status of a vacuum circuit breaker based on data fusion according to claim 5, characterized in that, The analysis of the distribution patterns of historical prediction errors and the construction of an error distribution model centered on the single predicted value specifically include: Collect historical data on the actual remaining life of vacuum circuit breakers of the same model under multiple different health indices, and the corresponding predicted values ​​obtained through the nonlinear mapping relationship model; Calculate the prediction error for each data point, where the prediction error is the difference between the actual remaining lifetime and the predicted remaining lifetime. Statistical analysis is performed on the prediction error of all data points to fit its probability distribution, which usually conforms to a normal distribution or a t-distribution with a mean of zero. The mean of the fitted error probability distribution is set to zero, and the variance is set to the sample variance of the historical prediction error, thereby constructing an error distribution model centered on the single prediction value and whose distribution shape is determined by the historical error.

7. A vacuum circuit breaker life condition assessment system based on data fusion, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the data fusion-based vacuum circuit breaker life status assessment method according to any one of claims 1 to 6.

Images