A switch cabinet temperature feature driven operating state evaluation method and system

Abstract

The application belongs to the technical field of data processing, and discloses a switch cabinet temperature feature driven operation state evaluation method and system, a system builds a double-layer discriminant model of working conditions and data distortion, original time sequence temperature data is divided into four categories of states through a sliding window variation coefficient discriminant method, adaptive wavelet double threshold filtering is used for random noise, a decomposition layer number and a threshold parameter are dynamically matched, noise is filtered out while weak fault features are completely retained, a composite filling algorithm of time sequence linear interpolation combined with working condition correction coefficient is used for working condition mutation data, data fluctuation deviation is corrected, and fault distortion data is directly marked and removed; a three-dimensional coupling feature mining system of time domain evolution, space distribution and working condition correlation is constructed, core features such as temperature rise rate, space temperature difference and working condition correlation degree are quantitatively extracted, and hidden hazards that cannot be identified by traditional threshold methods such as slow temperature rise, local overheating and periodic abnormalities are captured.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically a method and system for evaluating the operating status of switchgear driven by temperature characteristics. Background Technology

[0002] Switchgear is a core and critical piece of equipment in power transmission and distribution systems. Overheating faults in its core components, such as contacts, busbars, and terminals, are a major cause of equipment burnout, power outages, and grid fluctuations. Current switchgear operation status monitoring and evaluation technologies face the following technical challenges: Traditional monitoring and assessment methods rely on passive determination of raw temperature thresholds, which is a shallow data comparison process. They determine the equipment status only by comparing the real-time temperature value with a fixed threshold, without deeply exploring the temporal variation characteristics, spatial distribution characteristics, and temperature rise evolution characteristics of the temperature data. For example, hidden overheating hazards such as slow temperature rise, local temperature difference distortion, and periodic temperature anomalies cannot be identified. Traditional monitoring and assessment methods suffer from serious problems of underreporting and lag.

[0003] Switchgear operation is affected by multiple factors such as load fluctuations, ambient temperature and humidity, equipment aging, and closing and opening operations. The collected temperature time series data is mixed with a large amount of random noise, abrupt outliers and missing data. Traditional fixed filtering, mean imputation and other standardized data processing algorithms cannot achieve adaptive purification under operating conditions, resulting in the effective temperature characteristics being covered by noise.

[0004] Currently, mainstream assessment systems use algorithm architectures with fixed weights and fixed judgment rules. They cannot dynamically adjust the weights of temperature characteristics according to different cabinet types, different years of operation, and different load conditions. They also cannot distinguish between normal load temperature rise and abnormal fault temperature rise, which easily leads to false alarms. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for evaluating the operating status of switchgear driven by temperature characteristics, so as to solve one or more problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a switchgear temperature characteristic-driven operation status evaluation system, comprising the following modules: Furthermore, the operating condition data purification module collects real-time time-series temperature data from the switchgear temperature measurement nodes, and synchronously associates it with four types of operating condition parameters: load fluctuation status, ambient temperature and humidity parameters, equipment operating years, and opening and closing operation records. It builds a two-layer discrimination model for operating conditions and data distortion, and uses the sliding window coefficient of variation discrimination method to perform layered identification processing on the time-series temperature data. The collected raw data is divided into four data state types: normal fluctuation data, random noise data, operating condition change data, and fault distortion data. Different processing logic is configured for different state data. Normal fluctuation data is directly incorporated into a standardized dataset. For random noise data, an adaptive wavelet dual-threshold filtering algorithm is used to dynamically match the wavelet decomposition level and soft / hard threshold parameters based on the real-time operating condition fluctuation amplitude, filtering out data noise while retaining weak fault characteristic signals. For abrupt operating condition changes, a composite filling algorithm combining time-series linear interpolation and operating condition correction coefficients is used to correct data fluctuation deviations based on current equipment load and ambient temperature and humidity parameters. For fault-distorted data, real-time labeling and removal operations are performed, ultimately outputting a standardized temperature dataset.

[0007] Furthermore, the feature extraction and fusion module constructs a three-dimensional coupled feature mining system based on standardized time-series temperature data, which includes time-domain evolution, spatial distribution, and operating condition correlation. It deconstructs the fault-sensitive latent features in the temperature data and identifies potential equipment hazards based on multi-dimensional data parsing logic. The time-domain dimension is used to quantify and calculate parameters such as temperature rise rate, temperature drop gradient, temperature fluctuation variance, steady-state offset, frequency of periodic anomalies, and duration of exceeding thresholds, thereby capturing the changing trends of hidden faults such as slow temperature rise and intermittent temperature anomalies in the equipment. Spatial dimension measurement of parameters such as temperature difference at multiple nodes in the same cabinet, spatial unevenness, and global deviation of single-point temperature, to locate potential problems such as local overheating of equipment and temperature difference distortion in the temperature measurement area; The parameters of load and temperature correlation, environmental and temperature deviation coefficient, and operating condition temperature disturbance value are quantified in the working condition dimension. Invalid data fluctuations caused by environment, operation, and load are removed and valid fault feature information is locked. The feature extraction and fusion module simultaneously employs an improved extreme value normalization algorithm to unify the feature dimensions, dynamically adjusts the normalization calculation interval based on the equipment's operating years, generates a comprehensive feature matrix of switchgear temperature through feature fusion operations, retains various fault-sensitive core features, and realizes the characterization of hidden overheating hazards in switchgear based on multi-dimensional feature coupling logic.

[0008] Furthermore, the dynamic feature weighting module is configured with a three-level linkage dynamic weighting algorithm of basic entropy weight assignment, real-time working condition correction and life cycle decay fine-tuning. It achieves adaptive matching of feature weights through multi-level parameter calculation logic, calculates the state discrimination of temperature features in each dimension based on the information entropy algorithm combined with the fault sample dataset, and completes the basic entropy weight assignment according to the discrimination value. Four types of operating condition factors—real-time load level, ambient temperature, equipment runtime, and operation frequency—are introduced to construct an operating condition correction coefficient matrix, which adjusts the basic entropy weight value in real time to adapt to different equipment operating scenarios. The dynamic feature weighting module sets an equipment aging life cycle decay factor and dynamically adjusts the weight ratio of temperature features based on the equipment's operating years and historical fault frequency data. The three-level linkage dynamic weighting algorithm logic can distinguish between normal aging temperature rise, steady-state temperature rise under load, and abnormal temperature rise due to faults. It dynamically adjusts the weight ratio of various features according to the real-time operating scenario to amplify the identification of abnormal features.

[0009] Furthermore, the state coupling modeling module constructs a nonlinear coupling correlation model between multi-dimensional temperature features and equipment operating status. The nonlinear coupling correlation model takes the normalized multi-dimensional temperature feature vector as input and five operating statuses of the switchgear—optimal operation, normal stability, minor hidden danger, moderate warning, and severe fault—as output labels. Based on historical operating condition samples and fault sample datasets, a dynamic nonlinear correlation matrix is ​​generated through training. A polynomial piecewise fitting algorithm is used to quantify the influence weight of different temperature feature distortion amplitudes on equipment fault status and establish the correspondence between feature parameters and fault levels. The nonlinear coupling correlation model has built-in state recognition operation logic, which can identify the normal state of the equipment with small temperature fluctuations and parameters that do not exceed the standard but belong to the normal operating conditions. At the same time, it can capture the hidden fault state where the temperature parameters do not exceed the set threshold, but the multi-dimensional features are continuously distorted. Through multi-feature nonlinear coupling operation, it can lock in the hidden dangers of equipment abnormalities.

[0010] Furthermore, the operational status assessment module establishes a three-layer fuzzy clustering judgment logic: single-feature status preliminary judgment, multi-feature fusion verification, and operating condition calibration and classification. Based on hierarchical and progressive calculation rules, it completes the equipment operational status judgment. The single-feature status preliminary judgment combines the threshold range of various temperature features with real-time fluctuation status to complete the preliminary definition of equipment operational risks. The multi-feature fusion verification verifies the accuracy of the preliminary judgment result through the weighted fusion calculation result of multi-dimensional features. The operating condition calibration and classification combines the real-time operating conditions of the equipment and the equipment aging status parameters to complete the final classification of the equipment operational status level. The operational status assessment module ultimately outputs five levels of operational status results: optimal operation, normal and stable, minor hidden danger, moderate warning, and serious fault. It also generates confidence score data corresponding to the status level, presenting the degree of equipment operation risk in the form of quantitative data.

[0011] Furthermore, the prediction and source analysis module is equipped with a multi-feature prediction model based on improved temporal convolution. It performs temporal extrapolation calculations on fault features such as temperature rise rate, fluctuation variance, and spatial imbalance, so as to predict the evolution trend of equipment temperature features and identify potential risks such as increased equipment temperature rise and worsening feature distortion in advance. The predictive source analysis module is configured with a reverse feature association source tracing algorithm. After the equipment status is determined to be abnormal, the algorithm locks the temperature measurement node, core distortion features and abnormal causes corresponding to the abnormality through the operation logic of feature weight reverse deduction and operating condition factor reverse matching. It can distinguish four types of abnormal scenarios: equipment aging fault, load overload fault, environmental interference abnormality and operation disturbance abnormality, locate the root cause of the fault and output the abnormality cause and fault location information.

[0012] Furthermore, the model iteration optimization module establishes a hierarchical iteration optimization system for the entire life cycle of equipment, which includes classified storage, hierarchical training, and dynamic iteration. It collects temperature characteristic data, operating condition data, manual fault review results, and status judgment sample data in real time during the operation of the switchgear, and constructs a dynamic sample database for equipment. The dynamic sample database for equipment completes the classification, archiving, and storage of sample data according to the classification rules of operating condition fluctuation samples, equipment aging samples, historical misjudgment samples, and new fault samples. The model iterative optimization module is based on an incremental learning algorithm. It performs iterative optimization of operational parameters such as coupled model weights, feature judgment thresholds, and fuzzy clustering parameters in layers. It performs reverse training to correct the model's operating condition adaptation deviation for misjudged samples, updates the model's operating condition correction coefficients for sample data corresponding to equipment aging, seasonal changes, and load characteristic changes, and expands the system's fault feature reserve for new fault samples.

[0013] This invention also provides a method for evaluating the operating status of a switchgear driven by temperature characteristics. Based on the above system, it includes the following specific steps: During the working condition data purification stage, time-series temperature data of each temperature measurement node of the switchgear are collected simultaneously, as well as four types of working condition data: load, environment, equipment age, and opening / closing. A two-layer discrimination model of working condition and data distortion is built. The sliding window coefficient of variation discrimination method is used to classify the original data. Data purification is completed through adaptive wavelet double threshold filtering algorithm, working condition correction interpolation filling, and distorted data labeling and removal, and a standardized temperature dataset is output. In the multi-dimensional feature fusion stage, a three-dimensional coupled feature mining system of time domain, space and working condition is constructed to quantitatively extract core features of temperature evolution, spatial distribution and working condition correlation. The feature dimension differences are eliminated by dynamic normalization and fused to generate a comprehensive temperature feature matrix. In the feature dynamic weighting stage, a three-level linkage dynamic weighting algorithm is adopted. Based on the information entropy algorithm, the feature basic entropy weight is assigned. The weight parameters are finely adjusted in real time by combining the working condition correction coefficient matrix. The feature weight ratio of different service equipment is optimized by combining the equipment life cycle decay factor. In the state coupling modeling stage, a multi-feature nonlinear coupled state model is constructed. The multi-dimensional feature vector is used as input and the five-level equipment operating status is used as output. The dynamic correlation matrix is ​​trained based on the sample dataset. The fault influence weight of feature distortion is quantified by polynomial piecewise fitting to distinguish between pseudo-anomalies in the working condition and hidden equipment faults. In the graded status assessment stage, the equipment status is graded through a three-layer fuzzy clustering logic of single feature initial judgment, multi-feature fusion verification, and working condition calibration and grading, and five levels of operating status and corresponding confidence scores are output. In the source tracing and iterative optimization phase, a multi-feature temporal convolutional prediction model is constructed and anomaly reverse source tracing analysis is performed. The sample database and model core parameters are dynamically updated based on the incremental learning algorithm.

[0014] The beneficial effects of this invention are as follows: 1. This invention establishes a dual-layer discrimination model for operating conditions and data distortion. It divides the original time-series temperature data into four states using a sliding window coefficient of variation method, and sets corresponding processing strategies for different data types. For random noise, it employs adaptive wavelet dual-threshold filtering, dynamically matching the decomposition layer number and threshold parameters to filter out noise while fully preserving weak fault characteristics. For data with abrupt changes in operating conditions, it uses a composite filling algorithm combining time-series linear interpolation and operating condition correction coefficients to correct data fluctuation deviations. Fault-distorted data is directly marked and removed. This effectively isolates operating condition interference and data distortion, outputting a standardized temperature dataset.

[0015] 2. This invention constructs a three-dimensional coupled feature mining system encompassing temporal evolution, spatial distribution, and operating condition correlation. It quantitatively extracts core features such as temperature rise rate, spatial temperature difference, and operating condition correlation, capturing hidden dangers that traditional threshold methods cannot identify, such as slow temperature rise, local overheating, and periodic anomalies. Simultaneously, it employs a three-level linkage dynamic weighting algorithm, combining information entropy, operating condition correction coefficient, and life cycle decay factor to dynamically adjust feature weights. This clearly distinguishes between normal aging temperature rise, steady-state temperature rise under load, and abnormal temperature rise due to faults, amplifying the identification of abnormal features. It is adaptable to equipment with different service years and different load conditions, improving the accuracy of hidden fault identification and the reliability of assessment.

[0016] 3. The system of this invention classifies equipment status into five levels and outputs confidence scores through nonlinear coupling modeling and three-layer fuzzy clustering logic. It can identify normal fluctuations under normal operating conditions and also capture hidden faults caused by multi-dimensional feature distortion. Equipped with an improved temporal convolutional prediction model, it can predict the evolution trend of temperature features from 1 to 24 hours, realize early warning of faults, and the reverse tracing algorithm can quickly locate abnormal nodes and fault causes. At the same time, it builds a dynamic sample library of equipment based on incremental learning, and iteratively optimizes model parameters in layers to continuously adapt to scenarios such as equipment aging, seasonal changes, and new faults. Attached Figure Description

[0017] Figure 1 This is the overall flowchart for evaluating the operating status of the switchgear driven by temperature characteristics according to the present invention; Figure 2 This is a flowchart of the working condition data purification sub-process of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figures 1 to 2 As shown, this embodiment of the invention provides a switchgear temperature characteristic-driven operation status evaluation system, including the following modules: In this embodiment of the invention, the operating condition data purification module collects real-time time-series temperature data from the temperature measurement nodes of the switchgear, and synchronously associates it with four types of operating condition parameters: load fluctuation status, ambient temperature and humidity parameters, equipment operating years, and opening and closing operation records. It builds a two-layer discrimination model of operating condition and data distortion, and uses the sliding window coefficient of variation discrimination method to perform layered identification processing on the time-series temperature data. The collected raw data is divided into four data state types: normal fluctuation data, random noise data, operating condition change data, and fault distortion data. Different processing logic is configured for different state data, thereby completing the data standardization and purification processing. The sliding window coefficient of variation discrimination method is based on the continuous fluctuation characteristics of the temperature time series data of the switch cabinet. A fixed-length sliding window is set to traverse the time series data segment by segment. The window length is matched with the on-site temperature measurement sampling frequency and set to 20 consecutive sampling points. During the traversal, the dispersion and fluctuation pattern of the temperature data within the window are calculated in real time. Formula for calculating the coefficient of variation of a sliding window: , The coefficient of variation represents the time-series temperature data within the sliding window, used to quantify the dispersion and fluctuation of the temperature data; It represents the standard deviation of time-series temperature data within the sliding window, reflecting the discrete fluctuation state of the temperature data within the window; This represents the average value of the time-series temperature data within the sliding window, reflecting the overall level of the temperature data within the window.

[0020] Based on the calculation results, a four-level stratification threshold was set to divide the data into four states: data with dispersion in the normal range and no sudden rises or falls is normal fluctuation data; data with extremely small dispersion but random noise interference is random noise data; data with short-term large fluctuations caused by load jumps or sudden environmental changes is abnormal operating condition data; data with values ​​exceeding the physical temperature measurement range of the equipment and continuous abnormal jumps is fault distortion data. The sliding window adopts a point-by-point sliding traversal rule, and the window sliding step size is consistent with the interval of temperature sampling points. During the window traversal, the dispersion and fluctuation amplitude of the data are calculated in real time. Combined with the physical change range of switch cabinet temperature and the fluctuation law of normal operation, the status of the data is determined step by step.

[0021] The dual-layer discrimination model for operating conditions and data distortion consists of an upper-layer operating condition adaptive discrimination layer and a lower-layer data distortion detection layer working in series. The two layers have clear division of labor and mutual verification. The upper-layer operating condition discrimination layer prioritizes four types of parameters: load, ambient temperature and humidity, equipment operating years, and circuit breaker operation records. It first determines whether temperature fluctuations are caused by legitimate changes in operating conditions and filters out reasonable data fluctuations caused by normal operating conditions. After eliminating operating condition interference, the lower-layer distortion detection layer performs refined distortion judgment based on the calculation results of the sliding window variation coefficient to distinguish the four types of data states. The dual-layer discrimination logic avoids data misclassification caused by a single detection rule.

[0022] The normal fluctuation data is directly incorporated into the standardized dataset; the random noise data is filtered using an adaptive wavelet dual threshold filtering algorithm, which dynamically matches the wavelet decomposition level and soft and hard threshold parameters according to the real-time operating condition fluctuation amplitude, so as to filter out data noise while retaining weak fault feature signals in the data. The adaptive wavelet dual-threshold filtering algorithm prioritizes the selection of wavelet basis functions that are suitable for the switchgear temperature signal, and automatically matches the optimal wavelet decomposition level according to the real-time operating condition fluctuation amplitude. The algorithm adopts a high and low dual-threshold collaborative judgment mechanism. The low threshold is used to filter out small random noise, and the high threshold is used to retain weak fault features in the temperature signal. The threshold value is adaptively adjusted in real time according to the load change, ambient temperature and humidity and other operating condition parameters. During the filtering process, the wavelet coefficients are dynamically corrected to completely retain the weak feature signals corresponding to latent faults such as slow temperature rise and local micro-heating while eliminating random noise.

[0023] The wavelet basis function is preferentially selected from the db4 wavelet basis, which is suitable for stable temperature signals. This wavelet basis can match the changing characteristics of the switchgear temperature signal to the greatest extent. The dual threshold adopts a linkage adjustment mechanism. The low threshold fluctuates slightly with the intensity of random noise, while the high threshold is synchronously adapted with the intensity of fault characteristics. The high and low thresholds work together to eliminate noise without damaging weak fault characteristic signals such as slow temperature rise and local micro-heating.

[0024] For the abrupt change data of the operating conditions, a composite filling algorithm combining time-series linear interpolation and operating condition correction coefficient is adopted to correct the data fluctuation deviation by combining the current equipment load and environmental temperature and humidity parameters; for the fault distortion data, real-time marking and removal operations are performed, and finally a standardized temperature dataset is output.

[0025] The composite filling algorithm first uses time-series linear interpolation as a basis to initially fill in missing or distorted data caused by sudden changes in operating conditions, with the interpolation step size consistent with the original temperature measurement sampling interval. On this basis, an operating condition correction coefficient is introduced, which is linked in real time to parameters such as the current load level, ambient temperature and humidity, and the opening and closing operation status to dynamically correct the initial interpolation results and eliminate temperature deviations caused by sudden changes in operating conditions. After the filling is completed, the data is verified for rationality to ensure that the filled temperature values ​​conform to the equipment's operating rules and have no obvious conflicts with data from adjacent time periods.

[0026] Reasonableness verification is performed by comparing the fluctuation difference between the supplemented data and the data in adjacent time periods, as well as the temperature difference with other temperature measurement nodes in the same cabinet, to determine whether the value is within a reasonable range. If the supplemented data exceeds the normal fluctuation range, the system will automatically readjust the operating condition correction coefficient and supplement it a second time until the data matches the actual operating status of the equipment.

[0027] In this embodiment of the invention, the feature extraction and fusion module constructs a three-dimensional coupled feature mining system based on standardized time-series temperature data, which includes time-domain evolution, spatial distribution, and operating condition correlation. It deconstructs the fault-sensitive latent features in the temperature data and identifies potential equipment hazards based on multi-dimensional data parsing logic. The three-dimensional coupled feature mining system employs a complete process of parallel extraction, cross-validation, redundancy removal, and fusion normalization to achieve feature coupling. Temporal, spatial, and operational condition features are extracted simultaneously and independently. After extraction, cross-validation is performed to remove redundant information that repeatedly represents faults. The retained effective features are then standardized and integrated to form a comprehensive feature set. After three-dimensional coupling, it can cover all representations of latent faults such as slow temperature rise, localized overheating, and periodic anomalies.

[0028] The time-domain dimension is used to quantify and calculate parameters such as temperature rise rate, temperature drop gradient, temperature fluctuation variance, steady-state offset, frequency of periodic anomalies, and duration of exceeding thresholds, thereby capturing the changing trends of hidden faults such as slow temperature rise and intermittent temperature anomalies in the equipment. Spatial dimension measurement of parameters such as temperature difference at multiple nodes in the same cabinet, spatial unevenness, and global deviation of single-point temperature, to locate potential problems such as local overheating of equipment and temperature difference distortion in the temperature measurement area; The parameters of load and temperature correlation, environmental and temperature deviation coefficient, and operating condition temperature disturbance value are quantified in the working condition dimension. Invalid data fluctuations caused by environment, operation, and load are removed and valid fault feature information is locked. The feature extraction and fusion module simultaneously employs an improved extreme value normalization algorithm to unify the feature dimensions, dynamically adjusts the normalization calculation interval based on the equipment's operating years, generates a comprehensive feature matrix of switchgear temperature through feature fusion operations, retains various fault-sensitive core features, and realizes the characterization of switchgear overheating hidden dangers under conditions without fixed threshold constraints based on multi-dimensional feature coupling logic.

[0029] The improved extreme value normalization algorithm addresses the issue of large differences in the numerical range of different temperature characteristics by dynamically adjusting the normalization calculation interval based on the equipment's operating years. The interval is narrower for newly commissioned equipment and appropriately wider for aging equipment. The algorithm first reasonably truncates extreme outliers in the feature data to avoid interference from extreme values ​​in the normalization results, and then maps all feature data to a unified standard interval. After normalization, the feature data is smoothed to ensure a uniform distribution of the feature matrix values.

[0030] Extreme outlier truncation is based on the statistical distribution of historical characteristic data of the equipment. Extreme values ​​that exceed the normal distribution range are replaced with reasonable values ​​at the range boundary. The truncation rules are dynamically adjusted according to the equipment's operating years.

[0031] In this embodiment of the invention, the dynamic feature weighting module is configured with a three-level linkage dynamic weighting algorithm of basic entropy weight assignment, real-time working condition correction and life cycle decay fine-tuning. It achieves adaptive matching of feature weights through multi-level parameter calculation logic, calculates the state discrimination of temperature features in each dimension based on the information entropy algorithm combined with the fault sample dataset, and completes the basic entropy weight assignment according to the discrimination value. The three-level linkage dynamic weighting algorithm constructs a complete calculation system around three core dimensions: fault distinguishability of switchgear temperature characteristics, real-time operating condition adaptability, and equipment aging and degradation. Each level of weight adjustment is tailored to the actual operating characteristics of the switchgear. The entropy weighting stage quantifies the state distinguishability of each temperature characteristic using the principle of information entropy. The core calculation formula is: , The base entropy weight of the j-th temperature feature represents the sensitivity of this feature to the identification of equipment fault states. Indicates the sample sequence number; Indicates the total number of samples; This represents the percentage of the value of the j-th temperature feature in the i-th sample; The real-time operating condition correction phase introduces four core operating condition factors for the switchgear: load level, ambient temperature, equipment operating time, and operation frequency. A four-dimensional linear weighted correction coefficient matrix is ​​constructed, and the correction formula is as follows: , representing the weight value of the j-th temperature feature after operating condition correction; This represents the basic entropy weight of the j-th temperature feature; This represents the operating condition correction factor, with a value range of [0.8, 1.2], which is dynamically adjusted according to the load level, ambient temperature, equipment operating time, and operation frequency. The lifecycle degradation fine-tuning stage incorporates the aging patterns of switchgear equipment to set a degradation factor, calculated using the following formula: , Indicates the equipment's aging lifespan degradation factor; This indicates the actual operating years of the switchgear, in years; the final comprehensive weight. This represents the final comprehensive weight of the j-th temperature feature after three-level linkage adjustment; This represents the weight value of the j-th temperature feature after operating condition correction; Indicates the equipment's aging lifespan degradation factor; By adjusting the three-level linkage, the temperature characteristic weights of switchgear under different service years and operating conditions are accurately adapted, maximizing the identification of fault and abnormal characteristics.

[0032] A working condition correction coefficient matrix is ​​constructed by introducing four types of working condition factors: real-time load level, ambient temperature, equipment runtime, and operation frequency. The basic entropy weight value is adjusted in real time to adapt to different equipment operating scenarios such as light load steady state, heavy load fluctuation, and high temperature and humidity. The operating condition correction coefficient matrix is ​​constructed using a method of independent assignment of single factors and comprehensive weighting of multiple factors. Each type of operating condition factor is set with an adaptive adjustment range that matches the operating characteristics of the switchgear. The load factor is independently assigned according to three levels: heavy load, medium load, and light load. The ambient temperature factor is dynamically adjusted according to the ranges of normal temperature, high temperature, and low temperature. The equipment running time and operation frequency factors are segmented and adapted according to the cumulative values. The matrix will be automatically updated once per second based on the operating condition data collected in real time on site.

[0033] The dynamic feature weighting module sets an equipment aging life cycle decay factor and dynamically adjusts the weight ratio of temperature features based on the equipment's operating years and historical fault frequency data, thereby optimizing the feature recognition accuracy of equipment at different service cycles. The three-level linkage dynamic weighting algorithm logic can distinguish between normal aging temperature rise, steady-state temperature rise under load, and abnormal temperature rise due to faults. It dynamically adjusts the weight ratio of various features according to the real-time operating scenario, amplifies the identification of abnormal features, and reduces the probability of misjudgment in equipment operating status assessment.

[0034] In this embodiment of the invention, the state coupling modeling module constructs a nonlinear coupling correlation model between multi-dimensional temperature features and equipment operating status. The nonlinear coupling correlation model takes the normalized multi-dimensional temperature feature vector as input and five operating statuses of the switchgear—optimal operation, normal stability, minor hidden danger, moderate warning, and severe fault—as output labels. Based on historical operating condition samples and fault sample datasets, a dynamic nonlinear correlation matrix is ​​generated through training. A polynomial piecewise fitting algorithm is used to quantify the influence weight of different temperature feature distortion amplitudes on equipment fault status and establish the correspondence between feature parameters and fault levels. The polynomial piecewise fitting algorithm uses a third-order continuously differentiable piecewise function to accurately fit the gradient change law of temperature characteristic distortion of switchgear. The data is divided into three continuous intervals: 0-10%, 10%-30%, and 30%-100% according to the temperature characteristic distortion amplitude. Each interval independently fits the correlation between characteristic distortion and fault impact. The three segmented intervals are divided based on the actual fault patterns of the switchgear: 0-10% corresponds to minor fluctuations in the normal operation of the equipment, 10%-30% corresponds to characteristic distortions of potential early-stage problems, and 30%-100% corresponds to characteristic anomalies that exacerbate equipment faults. The interval boundary values ​​are statistically calibrated using tens of thousands of actual switchgear fault samples, which can accurately match the characteristic change patterns of the equipment throughout its entire lifecycle from normal operation to fault occurrence.

[0035] The nonlinear coupling correlation model takes a 1×M-dimensional normalized feature vector as input, where M is the total number of three core features of the switchgear: time-domain evolution, spatial distribution, and operating condition correlation. It covers all fault-sensitive features such as temperature rise rate, spatial temperature difference, and load correlation. The model output is a probability vector of five operating states. The probability values ​​are mapped to the 0-1 interval through the Softmax normalization function, and finally outputs five standard state labels for the switchgear: optimal operation, normal and stable, minor hidden danger, moderate warning, and serious fault. The model generates a dynamic nonlinear correlation matrix through joint training of historical operating condition samples and fault samples, which can accurately distinguish between pseudo-anomalies caused by operating condition fluctuations and hidden faults caused by equipment defects.

[0036] The nonlinear coupling correlation model has built-in state recognition operation logic, which can identify the normal state of the equipment with small temperature fluctuations and parameters that do not exceed the standard but belong to the normal operating conditions. At the same time, it can capture the hidden fault state where the temperature parameter does not exceed the set threshold but the multi-dimensional features are continuously distorted. By using multi-feature nonlinear coupling operation, it avoids the state deviation caused by single parameter judgment and locks in the hidden dangers of equipment abnormality.

[0037] In this embodiment of the invention, the operating status assessment module establishes a three-layer fuzzy clustering judgment logic: single-feature state initial judgment, multi-feature fusion verification, and operating condition calibration and classification. Based on the hierarchical and progressive calculation rules, it completes the equipment operating status judgment. The single-feature state initial judgment combines the threshold range of various temperature features with the real-time fluctuation status to complete the preliminary definition of equipment operating risk. The multi-feature fusion verification verifies the accuracy of the initial judgment result through the weighted fusion calculation result of multi-dimensional features and eliminates the judgment error caused by single feature fluctuation. The operating condition calibration and classification combines the real-time operating conditions of the equipment and the equipment aging status parameters to complete the final classification of the equipment operating status level. The three-layer fuzzy clustering decision is based on the fuzzy C-means (FCM) clustering algorithm. The fuzzy coefficient m is set to 2, and the number of cluster centers is fixed at 5, corresponding to the five operating states of the switchgear. The membership function adopts a Gaussian function adapted to the temperature feature distribution. The single-feature state initial judgment layer completes the preliminary definition of equipment operation risk based on the dynamic threshold range and real-time fluctuation amplitude of each temperature feature. The threshold is adaptively updated according to the historical data of the equipment. The multi-feature fusion verification layer performs cross-validation after weighted fusion of all temperature features to eliminate the judgment error caused by random fluctuation of single features. The operating condition calibration and classification layer combines parameters such as real-time load of the switchgear, ambient temperature and humidity, and equipment aging status to perform final calibration on the verification results. The algorithm iteration termination condition is set to the error value of two adjacent cluster centers ≤ 0.001. After the iteration is completed, the confidence score of each state level is automatically generated, and the score result is directly linked to the cluster membership degree.

[0038] Formula for calculating membership degree in fuzzy C-means clustering: , The membership degree of the i-th temperature feature sample to the j-th cluster center is the core parameter for determining the operating status classification of the switchgear. This represents the sample number of the temperature feature to be evaluated, used to distinguish different feature sample data. This represents the cluster center number, and each cluster center corresponds one-to-one with one of the five operating statuses of the switchgear. This represents the total number of cluster centers, with a fixed value of five, corresponding to five levels of status: optimal operation, normal and stable, minor hidden dangers, moderate warning, and serious failure. The distance between the i-th sample and the j-th cluster center represents the Euclidean distance, which measures the degree of matching between the sample and the corresponding running state. The Euclidean distance between the i-th sample and the k-th cluster center is used to perform normalization calculation of membership degree. This represents the fuzzy coefficient, with a fixed value of two, ensuring the smoothness and rationality of clustering decisions.

[0039] The threshold used for the initial judgment of a single feature state is not a fixed value, but a dynamic threshold automatically calculated based on the equipment's historical 30-day normal operation data as a statistical benchmark, combined with the current real-time load, ambient temperature and humidity, and equipment aging level. For different types of features such as temperature rise rate, temperature drop gradient, and spatial temperature difference, the system will generate independent and adaptive dynamic threshold ranges. The thresholds can automatically and flexibly shift with seasonal changes, load adjustments, and equipment aging, which can not only sensitively capture minor fault distortion features, but also effectively filter out false abnormal signals caused by fluctuations in normal operating conditions.

[0040] The operational status assessment module ultimately outputs five levels of operational status results: optimal operation, normal and stable, minor hidden danger, moderate warning, and serious fault. It also generates confidence score data corresponding to the status level, presenting the degree of equipment operation risk in the form of quantitative data.

[0041] The fuzzy clustering determination takes the weighted fusion result of multidimensional temperature features as the core input. First, it presets the initial cluster center according to the five-level operating state standard, and then adjusts the membership degree value of each feature sample through multiple iterative calculations until the clustering result is stable and converged. During the judgment process, real-time operating parameters and equipment aging status data are simultaneously accessed for final calibration, and pseudo-abnormal states caused by load fluctuations and environmental changes are automatically filtered out.

[0042] In this embodiment of the invention, the prediction and source analysis module is equipped with a multi-feature prediction model based on improved temporal convolution. It performs temporal extrapolation calculations on fault features such as temperature rise rate, fluctuation variance, and spatial imbalance, so as to predict the evolution trend of equipment temperature features in the next 1-24 hours, identify potential risks such as increased equipment temperature rise and worsening feature distortion in advance, and complete the early warning of equipment failure. The improved temporal convolutional multi-feature prediction model is designed specifically for the scenario of predicting the temporal features of switch cabinet temperature. The overall network adopts a serial cascaded structure, consisting of an input feature embedding layer, three causal convolutional layers, two dilated convolutional layers, a feature fusion layer, and a fully connected output layer connected in sequence. The gradient vanishing problem is avoided between the layers through residual connections. Residual connections directly superimpose shallow raw temperature features with deep convolutional extracted features in a cross-layer direct connection manner, without any activation function or convolution operation compression; for long-term temperature prediction scenarios, residual connections automatically filter redundant interference information in time series data and only transmit effective signals related to fault features. Causal convolutional layers are used to lock the temporal dependencies of temperature data, and dilated convolutional layers capture the long-term evolution of switch cabinet temperature by expanding the receptive field. The convolutional kernels are uniformly set to a temporal adaptation size of 3×1, and the dilation rate increases stepwise in a geometric sequence of 1, 2, and 4. The activation function is the ReLU function adapted to temperature numerical regression. The model input is a 1×N dimensional standardized temperature feature time series vector, where N is the number of equally spaced time series sampling points within 24 hours. The sampling interval matches the temperature measurement frequency of 5 minutes / time at the switch cabinet site. The output is a 1×24 dimensional future time period temperature feature prediction sequence, which can cover the full-time prediction needs from 1 to 24 hours. During the model training phase, the Adam adaptive optimizer was used, with an initial learning rate of 0.001 and a batch size of 32. The loss function selected was the mean squared error, which had the best accuracy in fitting temperature values. The historical temperature feature data of the device for 14 consecutive days was used as the basic training set. After each round of training, a 10% validation set was used for verification. After 100 iterations, the model loss value tended to stabilize and converged.

[0043] Mean squared error loss function for temporal convolutional prediction models: , The mean squared error loss value is used to measure the fitting accuracy between the predicted and actual temperature feature values ​​and is a core evaluation indicator for model training and optimization. This represents the total number of temperature feature prediction samples, providing a basic statistical basis for loss function calculation; The index represents the sequence number of the predicted sample, used to distinguish the temperature feature prediction data at different times; The actual observed value representing the i-th temperature feature is the baseline reference data for model prediction; The model prediction value representing the i-th temperature feature is the prediction result output by the improved temporal convolution model.

[0044] The predictive source analysis module is configured with a reverse feature association source tracing algorithm. After the equipment status is determined to be abnormal, the algorithm locks the temperature measurement node, core distortion features and abnormal causes corresponding to the abnormality through the operation logic of feature weight reverse deduction and operating condition factor reverse matching. It can distinguish four types of abnormal scenarios: equipment aging fault, load overload fault, environmental interference abnormality and operation disturbance abnormality, locate the root cause of the fault and output the abnormality cause and fault location information.

[0045] The reverse feature association tracing algorithm starts with the anomaly judgment result and performs reverse tracing according to the path of feature distortion, weight positioning, operating condition matching, node locking, and cause classification. First, the core distortion features that contribute the most to the anomaly judgment are selected. Then, the corresponding temperature measurement node and feature dimension are locked by reverse mapping of feature weights. Subsequently, real-time operating condition factors are matched in reverse to distinguish whether the temperature anomaly is caused by equipment aging, overload, environmental interference, or operational disturbance. The algorithm has built-in feature matching rules for four types of abnormal scenarios, which can quickly locate the specific location of the fault, the type of distortion feature and the direct cause, and output the abnormal source tracing results.

[0046] The feature contribution is calculated by the weight ratio of each feature in the anomaly judgment. The higher the contribution, the greater the impact of the feature on the anomaly result. After the system locks the core distortion features according to the contribution ranking, it directly maps them to the corresponding temperature measurement physical nodes to quickly locate the specific location of the fault.

[0047] Each of the four types of abnormal scenarios has its own unique feature matching rules. Equipment aging faults are judged by a continuously rapid rate of temperature rise and a slow increase in the temperature difference in the space. Overload faults are judged by a sudden temperature rise that occurs synchronously with the load and a synchronous increase in the fluctuation amplitude. Environmental interference anomalies are judged by a synchronous shift in temperature across all nodes and no local temperature difference changes. Operational disturbance anomalies are judged by a short-term temperature jump after a circuit breaker operation and a rapid return to stability. The type of anomaly can be quickly identified through these unique rules.

[0048] In this embodiment of the invention, the model iterative optimization module establishes a hierarchical iterative optimization system for the entire life cycle of equipment, which includes classified storage, hierarchical training, and dynamic iteration. It collects temperature characteristic data, operating condition data, manual fault review results, and status judgment sample data in real time during the operation of the switchgear, and constructs a dynamic sample database for equipment. The dynamic sample database for equipment completes the classification, archiving, and storage of sample data according to the classification rules of operating condition fluctuation samples, equipment aging samples, historical misjudgment samples, and new fault samples. The equipment lifecycle hierarchical iterative optimization system is divided into three independent levels based on the switchgear's commissioning time and operating status: the new commissioning adaptation layer, the stable operation optimization layer, and the aging equipment correction layer. The new commissioning adaptation layer aims to quickly collect initial data and calibrate basic model parameters, with a high update frequency. The stable operation optimization layer focuses on periodically correcting operating condition coefficients and optimizing feature weights to adapt to seasonal and load changes. The aging equipment correction layer focuses on strengthening the identification of aging-related fault characteristics and relaxing the reasonable temperature rise judgment range.

[0049] The model iterative optimization module is based on an incremental learning algorithm. It performs iterative optimization of operational parameters such as coupled model weights, feature judgment thresholds, and fuzzy clustering parameters in layers. It performs reverse training to correct the model's operating condition adaptation deviation for misjudged samples, updates the model's operating condition correction coefficients for sample data corresponding to equipment aging, seasonal changes, and load characteristic changes, and expands the system's fault feature reserve for new fault samples.

[0050] The incremental learning adopts a mode that combines batch incremental learning with real-time updates. It is designed specifically for the full life cycle operation characteristics of switchgear. The number of effective samples in a single batch of incremental learning is no less than 100. The sample selection prioritizes misjudged samples with a confidence level of <0.7, while also including samples from typical scenarios such as equipment aging, seasonal changes, and load characteristic changes. The model iteration step size is fixed at 0.005. After every 500 new valid samples, a full parameter update process is automatically triggered to simultaneously optimize the three core modules of coupled model weights, feature judgment thresholds, and fuzzy clustering parameters. The fault feature library expansion adopts a similarity judgment mechanism. When the cosine similarity between a new fault feature and the existing feature library is less than 0.8, it is judged as a completely new fault type and the feature library expansion is completed. For misjudged samples, a reverse training process is performed to correct the model's operating condition adaptation bias. For gradual changes such as season, load, and aging, the operating condition correction coefficient matrix is ​​updated regularly.

[0051] Temperature feature prediction, anomaly tracing analysis, and model iterative optimization work together. Temperature feature prediction anticipates feature change trends in advance, providing clear directional guidance for anomaly tracing. The fault nodes, core distortion features, and cause types identified by anomaly tracing are directly used as high-quality training samples for model iterative optimization. After model iterative optimization, the accuracy of feature prediction and anomaly tracing are improved simultaneously, continuously correcting judgment biases and enriching the fault feature database.

[0052] This invention also provides a method for evaluating the operating status of a switchgear driven by temperature characteristics. Based on the above system, it includes the following specific steps: During the working condition data purification stage, time-series temperature data of each temperature measurement node of the switchgear are collected simultaneously, as well as four types of working condition data: load, environment, equipment age, and opening / closing. A two-layer discrimination model of working condition and data distortion is built. The sliding window coefficient of variation discrimination method is used to classify the original data. Data purification is completed by adaptive wavelet double threshold filtering, working condition correction interpolation filling, and distorted data labeling and removal, and a standardized temperature dataset is output. In the multi-dimensional feature fusion stage, a three-dimensional coupled feature mining system of time domain, space and working condition is constructed to quantitatively extract core features of temperature evolution, spatial distribution and working condition correlation. The feature dimension differences are eliminated by dynamic normalization and fused to generate a comprehensive temperature feature matrix. In the feature dynamic weighting stage, a three-level linkage dynamic weighting algorithm is adopted. Based on the information entropy algorithm, the feature basic entropy weight is assigned. The weight parameters are finely adjusted in real time by combining the working condition correction coefficient matrix. The feature weight ratio of different service equipment is optimized by combining the equipment life cycle decay factor. In the state coupling modeling stage, a multi-feature nonlinear coupled state model is constructed. The multi-dimensional feature vector is used as input and the five-level equipment operating status is used as output. The dynamic correlation matrix is ​​trained based on the sample dataset. The fault influence weight of feature distortion is quantified by polynomial piecewise fitting to distinguish between pseudo-anomalies in the working condition and hidden equipment faults. In the graded status assessment stage, the equipment status is graded through a three-layer fuzzy clustering logic of single feature initial judgment, multi-feature fusion verification, and working condition calibration and grading, and five levels of operating status and corresponding confidence scores are output. In the source tracing and iterative optimization phase, a multi-feature temporal convolutional prediction model is constructed and anomaly reverse source tracing analysis is performed. The sample database and model core parameters are dynamically updated based on the incremental learning algorithm.

[0053] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A switchgear temperature characteristic-driven operation status evaluation system, characterized in that, Includes the following modules: The operating condition data purification module is used to collect time-series temperature data of switchgear, associate it with equipment operating condition parameters, perform state stratification and purification processing on the time-series temperature data, and generate a standardized temperature dataset. The feature extraction and fusion module is used to mine and extract multi-dimensional temperature features based on the standardized temperature dataset, and after completing feature normalization processing, fuse them to construct a comprehensive temperature feature matrix for the switchgear. The dynamic feature weighting module is used to dynamically assign weights to each temperature feature in the comprehensive temperature feature matrix and adjust the weight ratio of each temperature feature. The state coupling modeling module is used to construct a coupled correlation model between temperature characteristics and equipment operating status, and to perform nonlinear coupling calculations between multi-dimensional temperature characteristics and equipment operating status. The operation status assessment module is used to determine the equipment operation status classification based on the coupling calculation results and generate confidence scores for the corresponding five levels of operation status. The predictive source analysis module is used to deduce the evolution trend of equipment temperature characteristics, perform reverse source analysis on abnormal states, and locate abnormal nodes and abnormal scenarios. The model iteration and optimization module is used to collect equipment operation data to build a dynamic sample library, iteratively update the core parameters of the model, and expand the system fault feature reserve library.

2. The switchgear temperature characteristic-driven operation status evaluation system according to claim 1, characterized in that, The operating condition data purification module collects real-time time-series temperature data from the switchgear temperature measurement nodes, and synchronously associates it with four core operating condition parameters: load fluctuation status, ambient temperature and humidity parameters, equipment operating years, and opening and closing operation records. A two-layer discrimination model for operating conditions and data distortion is established, and a sliding window coefficient of variation discrimination method is used to perform layered identification processing on the time-series temperature data. The original data is divided into four state types: normal fluctuation data, random noise data, operating condition change data, and fault distortion data. Normal fluctuation data is directly included in the standardized dataset; random noise data is processed using an adaptive wavelet dual-threshold filtering algorithm; operating condition change data is processed using a composite filling algorithm combining time-series linear interpolation and operating condition correction coefficients; and fault distortion data is marked and removed in real time. Finally, a standardized temperature dataset is output.

3. The switchgear temperature characteristic-driven operation status evaluation system according to claim 2, characterized in that, The feature extraction and fusion module constructs a three-dimensional coupled feature mining system encompassing temporal evolution, spatial distribution, and operating condition correlation. In the temporal dimension, it quantifies and calculates parameters including temperature rise rate, temperature drop gradient, and temperature fluctuation variance. In the spatial dimension, it measures parameters including temperature difference between multiple nodes within the same cabinet and spatial imbalance. In the operating condition dimension, it quantifies parameters including load-temperature correlation and environmental-temperature offset coefficients. It removes invalid temperature fluctuations caused by operating condition factors, retains valid fault features, and dynamically adjusts the feature normalization calculation interval based on the equipment's service life. After completing the unified processing of feature dimensions, it fuses and generates a comprehensive temperature feature matrix for the switchgear.

4. The switchgear temperature characteristic-driven operation status evaluation system according to claim 3, characterized in that, The dynamic feature weighting module is configured with a three-level linkage dynamic weighting algorithm: basic entropy weight assignment, real-time operating condition correction, and life cycle decay fine-tuning. Based on the information entropy algorithm and combined with the fault sample dataset, the state distinguishability of temperature features in each dimension is calculated to complete the basic entropy weight assignment. Four types of operating condition factors, namely real-time load level, ambient temperature, equipment running time, and operation frequency, are introduced to construct an operating condition correction coefficient matrix to correct the basic entropy weight assignment value in real time. A life cycle decay factor based on equipment aging is set, and the weight ratio of each temperature feature is dynamically adjusted in combination with the equipment's operating years and historical fault frequency data to amplify the identification of abnormal features.

5. The switchgear temperature characteristic-driven operation status evaluation system according to claim 4, characterized in that, The state coupling modeling module builds a nonlinear coupling correlation model between multi-dimensional temperature features and equipment operating status. It takes the normalized multi-dimensional temperature feature vector as input and the five operating statuses of the switchgear as output labels: optimal operation, normal and stable, minor hidden danger, moderate warning, and serious fault. A dynamic nonlinear correlation matrix is ​​generated by training on historical operating condition samples and fault sample datasets. A multinomial piecewise fitting algorithm is used to quantify the influence weight of different temperature characteristic distortion amplitudes on equipment fault status. Potential abnormal hazards of equipment are identified through multi-feature nonlinear coupling operation.

6. The switchgear temperature characteristic-driven operation status evaluation system according to claim 5, characterized in that, The operational status assessment module establishes a three-layer fuzzy clustering judgment logic: initial judgment of single-feature status, verification of multi-feature fusion, and calibration and classification of operating conditions. The initial assessment of single-feature status combines the threshold range of various temperature features with real-time fluctuations to define the initial risk of operation. The multi-feature fusion verification verifies the initial assessment results through multi-dimensional feature weighted fusion calculation. The final assessment of the equipment's operating status is determined by combining the equipment's real-time operating conditions and aging status parameters. Finally, the five-level operating status results and the corresponding confidence scores for each status level are output.

7. The switchgear temperature characteristic-driven operation status evaluation system according to claim 6, characterized in that, The prediction and tracing analysis module is equipped with a multi-feature prediction model based on improved temporal convolution. It performs time-series extrapolation on fault features including temperature rise rate, fluctuation variance, and spatial imbalance to predict the evolution trend of equipment temperature features. It is configured with a reverse feature association tracing algorithm, which locks the temperature measurement node, core distortion features and abnormal causes corresponding to the anomaly through feature weight reverse extrapolation and reverse matching of operating conditions. It distinguishes four types of abnormal scenarios: equipment aging, overload, environmental interference, and operational disturbance, and outputs the corresponding abnormal causes and fault location information.

8. The switchgear temperature characteristic-driven operation status evaluation system according to claim 7, characterized in that, The model iterative optimization module establishes a hierarchical iterative optimization system for the entire equipment lifecycle. It collects temperature characteristic data, operating condition data, manual fault review results, and status judgment sample data in real time. These data are archived and stored according to classification rules for operating condition fluctuation samples, equipment aging samples, historical misjudged samples, and new fault samples, thus constructing a dynamic equipment sample database. Based on the incremental learning algorithm, it iteratively optimizes the coupled model weights, feature judgment thresholds, and fuzzy clustering parameters. It performs reverse training to correct the model's operating condition adaptation deviation for misjudged samples, updates the operating condition correction coefficients for samples of equipment aging, seasonal changes, and load characteristic changes, and expands the system's fault feature reserve for new fault samples.

9. A method for evaluating the operating status of a switchgear driven by temperature characteristics, based on the system described in any one of claims 1-8, characterized in that, The specific steps include the following: During the working condition data purification stage, time-series temperature data of each temperature measurement node of the switchgear are collected simultaneously, as well as four types of working condition data: load, environment, equipment age, and opening and closing. A two-layer discrimination model for working condition and data distortion is built to complete the classification and purification of the original time-series temperature data and output a standardized temperature dataset. In the multi-dimensional feature fusion stage, a three-dimensional coupled feature mining system of time domain, space and working condition is constructed to quantitatively extract multi-dimensional fault-sensitive core temperature features. The feature dimension differences are eliminated by dynamic normalization and the temperature comprehensive feature matrix is ​​generated by fusion. In the feature dynamic weighting stage, a three-level linkage dynamic weighting algorithm is adopted to complete the basic assignment of feature entropy weights. Combined with the operating condition correction coefficient and the equipment life cycle decay factor, the weight ratio of each temperature feature is dynamically optimized. In the state coupling modeling stage, a multi-feature nonlinear coupled state model is constructed, with multi-dimensional feature vectors as input and five levels of equipment operating status as output. A dynamic correlation matrix is ​​trained based on the sample dataset to distinguish between pseudo-anomalies in operating conditions and latent equipment faults. In the graded status assessment stage, the equipment status is graded through a three-layer fuzzy clustering logic of single feature initial judgment, multi-feature fusion verification, and working condition calibration and grading, and the five-level operating status of the equipment and the corresponding confidence score are output. In the source tracing and iterative optimization phase, a multi-feature temporal convolutional prediction model is constructed to deduce the feature evolution trend, anomaly reverse source tracing analysis is performed in real time, and the sample database and model core parameters are dynamically updated based on the incremental learning algorithm.

Images