Method and equipment for evaluating reliability of intelligent small-sized switch cabinet

Abstract

The invention discloses an intelligent small switch cabinet reliability evaluation method and device, and relates to the technical field of switch cabinet intelligentization, and the method comprises the following steps: collecting the operation state data of a switch cabinet, and calculating the first reliability evaluation parameter of the switch cabinet according to the operation state data; obtaining life cycle data of the switch cabinet and a second reliability evaluation parameter, and calculating a dynamic threshold range corresponding to the second reliability evaluation parameter through collaborative learning of the shared tree and the special tree and constraint consistency correction; and performing correlation analysis on the first reliability evaluation parameter and the dynamic threshold range to obtain a switch cabinet reliability evaluation result. The method is used for solving the problem that the dimension is one-sided in the traditional reliability evaluation of the small-sized switch cabinet.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent switch cabinet, more particularly, the present application relates to a reliability evaluation method and device for intelligent small switch cabinet. BACKGROUND

[0002] In modern power systems, intelligent small switch cabinets, as the core equipment of power distribution and control, are widely used in various substations, distribution rooms and industrial sites. The operation reliability of intelligent small switch cabinets is directly related to the stability and safety of power supply. With the development of intelligent and distributed power systems, higher requirements are put forward for the reliability evaluation of switch cabinets, and the drawbacks of traditional evaluation methods are increasingly prominent.

[0003] For example, the patent for an invention with the announcement number CN120233176A discloses a switch cabinet operation reliability evaluation method, which belongs to the field of switch cabinets. The method includes: collecting the rated voltage value and the rated current value of multiple groups of drive plug-ins and power modules in the target switch cabinet; performing electrical variable protection testing on each group of drive plug-ins in the target switch cabinet to obtain the electrical variable protection testing results of the corresponding drive plug-ins; performing over-temperature protection testing on each group of power modules in the target switch cabinet to obtain the over-temperature testing results of the corresponding power modules; performing open-phase protection testing on the target switch cabinet to obtain the open-phase testing results corresponding to the target switch cabinet; and comprehensively evaluating the reliability of the target switch cabinet based on all the testing results.

[0004] The above-mentioned technical solution has at least the following technical problems: Currently, the reliability evaluation method for small switch cabinets in substations focuses on a single parameter dimension, ignoring the influence of environmental factors and mechanical performance on the operation reliability of the equipment, resulting in a single evaluation parameter dimension, which cannot fully reflect the real operating conditions of the switch cabinet, thereby affecting the accuracy and reliability of the evaluation results. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a reliability evaluation method and device for intelligent small switch cabinet, which generates a multi-dimensional dynamic reliability evaluation result through the collaborative learning and constraint consistency correction of shared trees and special trees, combined with a coupling rule library, to solve the problem of one-sidedness in traditional switch cabinet reliability evaluation.

[0006] To achieve the above-mentioned purposes, the present application provides the following technical solutions: The application discloses a reliability evaluation method for an intelligent small switch cabinet, and comprises the following steps: collecting switch cabinet operation state data, calculating a first reliability evaluation parameter of the switch cabinet according to the operation state data, obtaining switch cabinet life cycle data and a second reliability evaluation parameter, and calculating a dynamic threshold range corresponding to the second reliability evaluation parameter through collaborative learning of a shared tree and a special tree and constraint consistency correction; and performing correlation analysis on the first reliability evaluation parameter and the dynamic threshold range to obtain a switch cabinet reliability evaluation result.

[0007] In a preferred embodiment, the first reliability evaluation parameter of the switch cabinet comprises a partial discharge phase distribution entropy, a closing and opening coil current waveform distortion rate, a dew point-surface temperature difference and an ultrasonic wave attenuation coefficient of an insulating pull rod.

[0008] In a preferred embodiment, the calculation of the dynamic threshold range corresponding to the second reliability evaluation parameter through collaborative learning of the shared tree and the special tree and constraint consistency correction comprises the following steps: performing feature structuring processing on the switch cabinet life cycle data and the second reliability evaluation parameter to generate a structured feature vector, the second reliability evaluation parameter being obtained through historical fault data; modeling the structured feature vector by using a collaborative learning algorithm of the shared tree and the special tree, generating a plurality of decision trees through multiple rounds of iterative training, and fusing prediction results of the shared tree and the special tree to output an initial threshold range of the second evaluation parameter; and performing constraint consistency correction on the output initial threshold range to obtain the dynamic threshold range.

[0009] In a preferred embodiment, the modeling of the structured feature vector by using the collaborative learning algorithm of the shared tree and the special tree comprises the following steps: processing the structured feature vector by using a shared tree model to extract probability distribution and importance of features, generating an initial range of the evaluation parameter through multiple rounds of iteration, and outputting to the special tree through a bidirectional feedback mechanism; and based on the initial range and the bidirectional feedback mechanism, performing detailed learning on specific abnormal conditions by using a special tree model to generate a discrimination rule and feed back to correct an uncovered feature area of the shared tree model.

[0010] In a preferred embodiment, the bidirectional feedback mechanism comprises the following steps: feeding back the output probability distribution and feature importance information to the special tree by using the shared tree model; determining key features and optimizing a splitting strategy based on the feedback of the shared tree by using the special tree model, performing detailed modeling on specific abnormal conditions to generate a discrimination rule; and feeding back local high-risk parameters and the discrimination rule identified by the special tree to the shared tree, and correcting an uncovered feature area in a global model and adjusting weights of related features by the shared tree.

[0011] In a preferred embodiment, the constraint consistency correction comprises basic constraint correction and correlation constraint correction.

[0012] In a preferred embodiment, the correlation analysis is analyzed by a coupling rule base, and the coupling rule base is constructed by: performing correlation analysis on the first reliability evaluation parameter, identifying the interaction relationship between the parameters; using a constraint Apriori algorithm to mine the association rules of the first reliability evaluation parameter, generating an initial association rule set; calculating the confidence of each rule in the association rule set, and filtering and optimizing based on the preset confidence threshold; and summarizing the optimized association rules to construct the coupling rule base.

[0013] In a preferred embodiment, the constraint Apriori algorithm is used to mine the association rules of the first reliability evaluation parameter, and specifically: based on a dynamic threshold reverse model, a priori constraint condition is introduced to remove combinations that do not meet the constraints; and in combination with the dynamic threshold reverse model, the threshold range of the first reliability evaluation parameter is dynamically adjusted according to the risk level and importance.

[0014] In a preferred embodiment, the switch cabinet reliability evaluation result includes a first evaluation result passing through the first reliability evaluation parameter and the dynamic threshold range, and a second evaluation result of the first reliability evaluation parameter and the coupling rule base.

[0015] An electronic device, characterized in that the electronic device comprises at least one processor, and a memory in communication connection with the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the intelligent small switch cabinet reliability evaluation method according to any one of claims 1 to 9.

[0016] The present application is based on the technical effects and advantages of an intelligent small switch cabinet reliability evaluation method and device: The present application is based on the technical effects and advantages of an intelligent small switch cabinet reliability evaluation method and device: The present application is based on the technical effects and advantages of an intelligent small switch cabinet reliability evaluation method and device:

[0017] Figure 1 An intelligent small switch cabinet reliability evaluation method flowchart is provided for the embodiments of the present application.

[0018] Figure 2An electronic equipment structure schematic diagram of a smart small switch cabinet reliability evaluation equipment is provided in the embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0020] Embodiment 1, Figure 1 A smart small switch cabinet reliability evaluation method is provided in the present application, including the following steps: S1, collecting switch cabinet operation state data, and calculating a first reliability evaluation parameter of the switch cabinet according to the operation state data; S2, obtaining switch cabinet life cycle data and a second reliability evaluation parameter, and calculating a dynamic threshold range corresponding to the second reliability evaluation parameter through collaborative learning and constraint consistency correction of a shared tree and a special tree; S3, performing correlation analysis on the first reliability evaluation parameter and the dynamic threshold range, to obtain a switch cabinet reliability evaluation result.

[0021] In the embodiment, the switch cabinet life cycle data and the second reliability evaluation parameter are obtained, the dynamic threshold range corresponding to the second reliability evaluation parameter is calculated through collaborative learning and constraint consistency correction of a shared tree and a special tree, and the correlation between the first reliability evaluation parameters is evaluated in combination with a coupling rule library, to generate a final result. The core innovation is that the collaborative learning modeling of the shared tree and the special tree is performed, and the dynamic threshold range is obtained through constraint consistency correction, the actual failure boundary is deduced from the second evaluation parameter of the switch cabinet life cycle data and historical failure cases, and the dynamic threshold range is output, so that a personalized dynamic threshold range changing with the state and environment of each switch cabinet is generated. The coupling rule library adds deep combination of the structural characteristics of the small switch cabinet through correlation analysis of multiple parameters, and the evaluation result is improved. The problem of one-sidedness in the traditional reliability evaluation is solved.

[0022] S1, collecting switch cabinet operation state data, and calculating a first reliability evaluation parameter of the switch cabinet according to the operation state data; In the embodiment, the first reliability evaluation parameter of the switch cabinet includes a local discharge phase distribution entropy, a closing and opening coil current waveform distortion rate, a dew point-surface temperature difference, and an ultrasonic wave attenuation coefficient of an insulation pull rod.

[0023] It should be noted that the partial discharge phase distribution entropy is obtained by installing a special high-sensitivity sensor near the bus of the small switch cabinet, collecting the partial discharge pulse of the voltage cycle, counting the pulse number of each phase, and calculating the formula is:

[0024] In the formula, pi is the pulse proportion of the i-th phase interval, and H is the distribution entropy (normal range <1.2, aging >1.8); It should be noted that the distortion rate of the closing and opening coil current waveform is obtained by a Hall current sensor connected in series with the closing and opening coil loop of the small switch cabinet, collecting the current waveform when the closing and opening action is generated, and respectively performing Fourier transform on the collected current waveform and the standard current waveform to obtain the respective frequency spectrum representation. The distortion rate is obtained by calculating the difference between the two frequency spectrums, and the specific formula is:

[0025] In the formula, is the n-th harmonic current, is the fundamental wave, and TND is the output total harmonic distortion rate (normal <3%, mechanical abnormality >5%); It should be noted that the dew point-surface temperature difference is obtained by integrating the dew point sensor and the infrared temperature sensor in the small switch cabinet, and the difference between the dew point temperature and the bus surface temperature (normal >2℃, condensation risk ≤0℃) is calculated in real time. It should be noted that the ultrasonic wave attenuation coefficient of the insulating pull rod is obtained by the ultrasonic wave probes adsorbed on both ends of the pull rod of the small switch cabinet, injecting ultrasonic waves at the sending end, and detecting the amplitude attenuation at the receiving end, and the specific formula is:

[0026] In the formula, is the transmission amplitude, is the receiving amplitude, is the attenuation coefficient (normal <5%, crack >20%).

[0027] S2, obtain the switch cabinet life cycle data and the second reliability evaluation parameter, calculate the dynamic threshold range corresponding to the second reliability evaluation parameter through collaborative learning and constraint consistency correction of the shared tree and the special tree; In the embodiment, the dynamic threshold range corresponding to the second reliability evaluation parameter is calculated through collaborative learning and constraint consistency correction of the shared tree and the special tree, and specifically: The switch cabinet life cycle data and the second reliability evaluation parameter are subjected to feature structuring processing to generate a structured feature vector, and the second reliability evaluation parameter is obtained through historical fault data; The structured feature vector is modeled by using a shared tree and a special tree collaborative learning algorithm, a plurality of decision trees are generated through multiple rounds of iterative training, and the prediction results of the shared tree and the special tree are fused to output an initial threshold range of the second evaluation parameter; The output initial threshold range is constrained and consistent, and a dynamic threshold range is obtained.

[0028] It should be noted that the switchgear life cycle data includes: initial value of factory insulation resistance, local discharge quantity reference value, running time, cumulative on-off times, annual load rate curve; It should be noted that the second reliability evaluation parameter refers to the local discharge phase distribution entropy, the on-off coil current waveform distortion rate, the dew point-surface temperature difference and the ultrasonic attenuation coefficient of the insulation pull rod when the historical case fails. The abnormal values of the four parameters.

[0029] In this embodiment, the feature structuring process is specifically: The life cycle data and the second reliability evaluation parameter are divided into different dimensions according to features, and the dimensions include: basic dimension, time sequence dimension and environment dimension; The features are processed based on the Z-score standardization method to form a standardized feature row vector corresponding to the evaluation parameter.

[0030] It should be noted that the division into different dimensions is according to the specific type of the parameter. The initial value of the factory insulation resistance and the local discharge quantity reference value are divided into the basic dimension. The parameters that change with the running time, such as the cumulative on-off times and the running time, are divided into the time sequence dimension. The parameters related to the external environment are divided into the environment dimension.

[0031] It should be noted that the Z-score standardization of the features can eliminate the differences in dimensions and orders of magnitude, so that the different parameters can be collaboratively learned in the same model. The specific formula is:

[0032] In the formula, is the original feature value, is the mean value of the feature in the sample, is the standard deviation of the feature.

[0033] It should be noted that the standard deviation is calculated according to the formula:

[0034] In the formula, is the standard deviation of the feature, is the sample size of the feature, is the th sample of the feature. A raw sample value.

[0035] In this embodiment, the method for collaborative learning and constraint consistency correction of the shared tree and the special tree comprises a dynamic threshold back-propagation model.

[0036] In this embodiment, the structured feature vector is modeled using the collaborative learning algorithm of the shared tree and the special tree, specifically: The structured feature vector is processed using the shared tree model to extract the probability distribution and importance of the features, and after multiple iterations, an initial range of evaluation parameters is generated and output to the special tree through a bidirectional feedback mechanism; Using the special tree model, based on the initial range and the bidirectional feedback mechanism, the specific abnormal situation is refined and learned to generate a discrimination rule, and the uncovered feature area of the shared tree model is fed back and corrected.

[0037] In this embodiment, the process of shared tree learning modeling is specifically: Through the CART algorithm, at each round of splitting, all features and their possible splitting points are traversed, the Gini index of each feature at a given splitting point is calculated, and the feature with the smallest Gini index is selected for splitting; After selecting the best feature and splitting point, the data is recursively split into two subsets. Each subset continues to split until the stopping condition is reached (e.g., the tree depth reaches the preset maximum value, or the node purity reaches the threshold); At each leaf node, the probability distribution of the corresponding class is output; this probability distribution represents the probability of the system occurring a certain risk under the condition of the feature corresponding to the node.

[0038] It should be noted that the CART algorithm is specifically: According to the main features of the structured feature vector, the input parameters are divided into two parts; The mean square error is used as the purity measurement index to calculate the purity of the divided sample groups, and the sample groups are recursively split based on the purity calculation results; When the heterogeneity of the split sample group is lower than the preset heterogeneity threshold, or the number of samples in the sample group is less than the preset threshold, the splitting is stopped; The average value of the parameters of each sub-sample group obtained by the final division is calculated, and the average value is output as the threshold value of the corresponding parameter.

[0039] It should be noted that the specific formula for selecting the best feature and splitting point is:

[0040] In the formula, is the proportion of samples of the class, the smaller the Gini index, the purer the node.

[0041] In this embodiment, the process of learning and modeling the specialty tree is as follows: The split is performed by selecting the features most relevant to the output of the shared tree. The specialized tree will pay more attention to the boundary values ​​of these features, especially in areas with higher risk. Continue recursively splitting, selecting the best features and split points each time to maximize the purity after splitting; At the leaf nodes of the specialization tree, more detailed risk assessment results are output; the final output is a refined risk assessment result for specific features.

[0042] It should be noted that the specific formula for the risk assessment results output by integrating the shared tree and the specialized tree is as follows:

[0043] In the formula, This indicates the proportion of the shared tree in the final risk assessment. This represents the probability of risk assessment based on global features. This represents the probability of risk assessment based on local features.

[0044] In this embodiment, the bidirectional feedback mechanism is specifically as follows: The shared tree model is used to feed back the output probability distribution and feature importance information to the specialized tree; By using a specialized tree model, key features are determined based on shared tree feedback and the splitting strategy is optimized. Detailed modeling is then performed on specific anomalies to generate discrimination rules. The local high-risk parameters and their discrimination rules identified by the specialized tree are fed back to the shared tree. The shared tree corrects the feature regions not covered in the global model and adjusts the weights of the relevant features.

[0045] In this embodiment, before the shared tree and specialized tree learn together, Z-score preprocessing is required for all samples to calculate the initial sample weights. The calculation formula is:

[0046] In the formula, The baseline weighting coefficient represents the basic weight without other adjustment factors. Let i be the measure of uncertainty for sample i. Regarding uncertainty The mapping function is used to transform uncertainty into weight correction amounts. This is an indicator of the rarity or representativeness of sample i, reflecting the scarcity of the sample within the overall data distribution. , These are the weighting coefficients for uncertainty correction and scarcity correction, respectively. Let i be the observation value of sample i in feature dimension k. The dynamic threshold center position of feature k. This is a function of how much a sample deviates from the threshold, used to reflect the contribution of a sample to the learning of the threshold boundary. The weighting coefficient is used to correct for threshold deviation.

[0047] In this embodiment, the process of feeding back the shared tree to the specialized tree and the specialized tree determining key features and optimizing the splitting strategy based on the feedback information from the shared tree is as follows: The shared tree outputs the probability distribution of samples under each risk category and feeds this probability distribution back to the specialized tree as a soft label. When receiving sample input, the specialized tree uses this probability distribution along with the original feature vector of the sample as augmented input, thereby obtaining a smoother supervision effect during training and determining key features and optimizing the splitting strategy accordingly.

[0048] After receiving feedback suggestions from multiple special projects, the shared tree incorporates the rules provided by each project into the split search space as candidate split points using a candidate splitting method. For each candidate split *s*, a score is defined to characterize the split's discriminative power on the corrected samples. Splits with higher scores are selected first.

[0049] It should be noted that the formula for the candidate splitting alternative method is as follows:

[0050] in: This is used to measure the contribution of partition S to the improvement of node purity; This represents a confidence measure considering parameter correlation.

[0051] It should be noted that the above The definition is as follows:

[0052] In the formula, The sample set after correlation correction. The weights of sample i, To partition the set of samples that fall into the left child node L, Let i be an indicator function, if sample i belongs to If the value is 1, then the value is 1; otherwise, it is 0.

[0053] In this embodiment, the process of the shared tree correcting the uncovered feature regions in the global modeling and adjusting the weights of the partitioned features in the model based on feedback from the specialized tree is as follows: For each specialized tree, high-risk nodes that meet certain conditions are identified in its leaf nodes. These conditions include: the number of samples in the leaf node is not less than a preset threshold, the high-risk probability exceeds a preset value, and the classification confidence is high. For leaf nodes that meet the conditions, the discrimination rules determined by the split path are extracted, and the node statistics are recorded to form a correction sample set. The correction samples are assigned higher weights and fed back to the shared tree to correct feature regions that are not covered or under-learned during the modeling process. This further optimizes the weight allocation of the shared tree in feature selection and boundary determination, thereby improving the overall accuracy and robustness of the model. It should be noted that when the decision tree selects "waveform distortion rate of opening and closing coil > 12%" and "dew point temperature difference < 2℃" as splitting conditions in turn during the training process, all samples falling into the leaf node meet the above conditions. Therefore, the discrimination rule corresponding to the leaf node is "when the waveform distortion rate of opening and closing coil is greater than 12% and the dew point temperature difference is less than 2℃, it is judged as high risk".

[0054] In this embodiment, the constraint consistency correction includes basic constraint correction and associated constraint correction.

[0055] In this embodiment, the basic constraints are defined based on electrical industry standards and equipment operation specifications, setting initial upper and lower limits for the evaluation parameter thresholds. The correlation constraint correction is constructed based on the correlation between the second reliability assessment parameters and the physical characteristics of the switchgear.

[0056] It should be noted that for partial discharge phase distribution entropy, the limit for partial discharge in conventional switchgear is 10 pC, corresponding to an upper limit of 2.5 for phase distribution entropy; however, due to the lower insulation margin of small switchgear (insulation distance is 20% shorter than conventional switchgear), the threshold is tightened by 20%, hence the associated constraint is: = 2.0, = 0.3 (lower limit to avoid misjudging normal micro-discharges); It should be noted that the upper limit for the distortion rate of the opening and closing coils under normal mechanical operation is 10%; however, due to the compact structure of small switchgear and accelerated wear, the threshold is tightened to 10%, hence the associated constraint is: = 9.0%, =0.5% (lower limit excludes sensor zero drift); It should be noted that, considering the vibration amplification effect of the small cabinet (vibration value is 30% higher than that of the conventional cabinet), the risk of cracking is higher due to the attenuation coefficient of the insulating tie rod. Therefore, the associated constraint is: = 18% (10% lower than the standard cabinet's 20%); It should be noted that, regarding localized temperature differences within the cabinet, the risk of localized overheating is increased due to the limited heat dissipation of small cabinets (the heat dissipation area is only 60% of that of regular cabinets). Therefore, the associated constraints are as follows: =7℃ (2℃ lower than the standard cabinet temperature of 9℃).

[0057] S3, perform a correlation analysis on the first reliability assessment parameter and the dynamic threshold range to obtain the switch cabinet reliability assessment result.

[0058] In this embodiment, the correlation analysis is performed using a coupling rule base, which is constructed as follows: Correlation analysis was performed on the first reliability assessment parameters to identify the interaction relationships between the parameters; A constrained Apriori algorithm is used to mine association rules for the first reliability assessment parameter to generate an initial set of association rules. Calculate the confidence score of each rule in the association rule set, and perform filtering and optimization based on a preset confidence threshold; Summarize and optimize the association rules to build a coupled rule library.

[0059] In this embodiment, the process of performing correlation analysis on the first reliability assessment parameter and identifying the interaction relationship between the parameters is as follows: The parameter threshold range generated by the dynamic threshold backpropagation model is discretized and divided into insufficient, normal, and excessive levels according to the upper and lower limits of the threshold. The discretized levels correspond to different risk levels and are used as atomic terms of the Apriori algorithm. Based on the correction of association constraints, the feasibility of atomic terms is determined. If an atomic term does not meet the physical boundary conditions, it is directly eliminated and does not enter the candidate set. During the candidate generation process, the association constraint predicate of the candidate combination is calculated. If the candidate does not meet the predicate, it is directly discarded.

[0060] It should be noted that the predicate is composed of causal relationships or empirical formulas between parameters, such as "the insulation resistance shall not be greater than 200MΩ when the temperature exceeds 80℃".

[0061] It should be noted that the generated parameter threshold range is discretized using the following formula:

[0062] In the formula, and To evaluate the dynamic threshold range of the parameters .

[0063] It should be noted that during the rule filtering stage, a predetermined confidence threshold is set, and only rules with a confidence level greater than the threshold are retained, ensuring that the rules entering the rule base have high reliability and universality.

[0064] It should be noted that the confidence score calculation process for the combined rules is as follows:

[0065] In the formula, Indicates the condition for combining parameters. This indicates the corresponding risk event.

[0066] In this embodiment, the step of using the constrained Apriori algorithm to mine association rules for the first reliability assessment parameter specifically involves: Based on the dynamic threshold back-calculation model, prior constraints are introduced to remove combinations that do not meet the constraints. By combining a dynamic threshold back-calculation model, the threshold range of the first reliability assessment parameter is dynamically adjusted according to the risk level and importance.

[0067] In this embodiment, the process of dynamically adjusting the threshold range of the first reliability assessment parameter according to the risk level and importance is as follows: By using a dynamic threshold back-inference model to generate a threshold range, when the system risk level increases, the threshold range of the evaluation parameters is tightened, making the parameters more likely to trigger "anomaly" judgments, thereby enhancing the sensitivity to potential faults; when the risk level is low, the threshold range is widened to avoid generating too many false alarms.

[0068] It should be noted that the formula for this process is as follows:

[0069] In the formula, and The corrected threshold. This is the risk level coefficient. and This is an adjustment factor used to control the degree of threshold tightening.

[0070] In this embodiment, to verify the effectiveness of the method of the present invention, four small switchgear with different faults were selected as evaluation objects. The status and environmental data of the switchgear during operation were collected respectively, the evaluation parameter values ​​were calculated, and an evaluation was performed in combination with dynamic thresholds to generate evaluation results. The evaluation results are shown in Table 1: Table 1

[0071] As can be seen from Table 1, the method of the present invention can achieve a one-time evaluation of evaluation parameters and dynamic thresholds for small switchgear under different operating conditions, laying the foundation for further evaluation. For example, in Case 1, because the waveform distortion rate of the opening and closing coil current exceeded 5% and the dew point-surface temperature difference was below the threshold, the coil current abnormality risk was determined to be high and the condensation risk to be medium in a single assessment. In Case 3, the waveform distortion rate of the opening and closing coil current exceeded 2%, and the entropy of the partial discharge phase distribution exceeded the standard. The time difference between the two exceeding the standard was less than 10 minutes. In one assessment, the coil current abnormality risk was determined to be medium and the partial discharge risk was also medium. Case 4 was determined to have a medium risk of partial discharge and a medium risk of damage to the insulating tie rod due to excessive entropy of partial discharge phase distribution and ultrasonic attenuation coefficient of the insulating tie rod in a single assessment.

[0072] In this embodiment, the initial evaluation results are combined with the coupling rule base to perform correlation analysis on the parameters of the results and perform a second evaluation. The second evaluation results are shown in Table 2.

[0073] Table 2

[0074] As can be seen from Table 2, the method of the present invention can accurately determine the risk level of different switch cabinets through the joint analysis of dynamic thresholds and rule bases, and give the cause of risk, providing a reliable basis for subsequent operation and maintenance. For example, in Case 1, the first assessment determined that the risk of abnormal coil current was high and the risk of condensation was medium. By combining the coupling rule base, it was found that there was a rule that high current distortion rate + low dew point temperature difference -> high risk of coil contact oxidation. The generated second assessment result was high risk. The risk cause analysis was that poor contact of coil contacts and condensation formed a vicious cycle of "poor contact - condensation corrosion". In addition, the humidity accumulation in the compact space of the small switch cabinet further amplified the impact. In Case 2, the initial assessment revealed three risk scenarios: moderate risk of partial discharge, moderate risk of condensation, and moderate risk of damage to the insulating tie rod. The rule base identified a rule: high phase distribution entropy + low dew point temperature difference + high ultrasonic attenuation coefficient -> extremely high risk of insulation breakdown. Given the severity of the problem, the second assessment would assess the risk level as extremely high. The risk cause analysis revealed a triple coupling of "discharge-condensation-structural damage," with concentrated electric fields in a compact space leading to short-term failure of the insulation system. In Case 4, based on the results of a single assessment that the risk of partial discharge and the risk of damage to the insulating tie rod are both medium, and combined with the association of the coupling rule base, there is a rule that high phase distribution entropy + high ultrasonic attenuation coefficient -> significant risk of insulation tie rod breakage, with a risk level of high. The risk cause analysis is that partial discharge and physical damage to the insulating tie rod form a "discharge degradation - structural damage" coupling cycle, and the compact space of the small switch cabinet accelerates the insulation failure.

[0075] Example 2, Figure 2 A schematic diagram of the electronic device structure of the intelligent small switchgear reliability assessment method of the present invention is provided. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the intelligent small switchgear reliability assessment method.

[0076] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0077] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0078] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0079] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0080] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0081] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A reliability assessment method for intelligent small switchgear, characterized in that, Includes the following steps: Collect switchgear operating status data and calculate the first reliability assessment parameters of the switchgear based on the operating status data; The switchgear lifecycle data and second reliability assessment parameters are obtained. Through collaborative learning of shared tree and special tree and constraint consistency correction, the dynamic threshold range corresponding to the second reliability assessment parameters is calculated. A correlation analysis was performed on the first reliability assessment parameter and the dynamic threshold range to obtain the switchgear reliability assessment result.

2. The intelligent miniature switchgear reliability assessment method according to claim 1, characterized in that, The first reliability assessment parameters of the switchgear include: partial discharge phase distribution entropy, current waveform distortion rate of opening and closing coils, dew point-surface temperature difference, and ultrasonic attenuation coefficient of insulating tie rod.

3. The intelligent miniature switchgear reliability assessment method according to claim 2, characterized in that, The dynamic threshold range corresponding to the second reliability assessment parameter is calculated through collaborative learning of the shared tree and the specialized tree, as well as constraint consistency correction. Specifically: The switchgear lifecycle data and the second reliability assessment parameters are processed by feature structuring to generate a structured feature vector. The second reliability assessment parameters are obtained through historical fault data. The structured feature vector is modeled using a shared tree and specialized tree collaborative learning algorithm. Multiple decision trees are generated through multiple rounds of iterative training. The prediction results of the shared tree and specialized tree are then fused to output the initial threshold range of the second evaluation parameter. Constraint consistency correction is applied to the initial threshold range of the output to obtain the dynamic threshold range.

4. The intelligent miniature switchgear reliability assessment method according to claim 3, characterized in that, The method of modeling structured feature vectors using a shared tree and specialized tree collaborative learning algorithm is as follows: The shared tree model is used to process structured feature vectors, extract the probability distribution and importance of features, generate the initial range of evaluation parameters through multiple iterations, and output them to the specialization tree through a two-way feedback mechanism; By utilizing a specialized tree model, based on an initial range and a two-way feedback mechanism, we can refine the learning for specific anomalies, generate discrimination rules, and provide feedback to correct the uncovered feature regions of the shared tree model.

5. The intelligent miniature switchgear reliability assessment method according to claim 4, characterized in that, The two-way feedback mechanism is specifically as follows: The shared tree model is used to feed back the output probability distribution and feature importance information to the specialized tree; By using a specialized tree model, key features are determined based on shared tree feedback and the splitting strategy is optimized. Detailed modeling is then performed on specific anomalies to generate discrimination rules. The local high-risk parameters and their discrimination rules identified by the specialized tree are fed back to the shared tree. The shared tree corrects the feature regions not covered in the global model and adjusts the weights of the relevant features.

6. The intelligent miniature switchgear reliability assessment method according to claim 5, characterized in that, The constraint consistency correction includes basic constraint correction and associated constraint correction.

7. The intelligent miniature switchgear reliability assessment method according to claim 6, characterized in that, The correlation analysis is performed using a coupling rule base, which is constructed as follows: Correlation analysis was performed on the first reliability assessment parameters to identify the interaction relationships between the parameters; A constrained Apriori algorithm is used to mine association rules for the first reliability assessment parameter to generate an initial set of association rules. Calculate the confidence score of each rule in the association rule set, and perform filtering and optimization based on a preset confidence threshold; Summarize and optimize the association rules to build a coupled rule library.

8. The intelligent miniature switchgear reliability assessment method according to claim 7, characterized in that, The process of using a constrained Apriori algorithm to mine association rules for the first reliability assessment parameter is as follows: Based on the dynamic threshold back-calculation model, prior constraints are introduced to remove combinations that do not meet the constraints. By combining a dynamic threshold back-calculation model, the threshold range of the first reliability assessment parameter is dynamically adjusted according to the risk level and importance.

9. The intelligent miniature switchgear reliability assessment method according to claim 8, characterized in that, The switchgear reliability assessment results include a first assessment result based on a first reliability assessment parameter and a dynamic threshold range, and a second assessment result based on the first reliability assessment parameter and a coupled rule base.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the intelligent small switch cabinet reliability assessment method as described in any one of claims 1 to 9.

Images