Method and system for monitoring abnormal state of zinc oxide surge arrester in power distribution line

Abstract

The application provides an abnormal state monitoring method and system of zinc oxide arresters in power distribution lines, and relates to the technical field of measuring power transformation, which comprises the following steps: monitoring the zinc oxide arrester group of the power distribution lines in real time; performing multi-dimensional conduction accompanying fault prediction and multi-dimensional non-conduction accompanying fault prediction on the zinc oxide arrester group under balanced confrontation optimization according to the arrester monitoring data block; performing neighborhood action intervention compensation on the conduction accompanying fault graph and the non-conduction accompanying fault graph according to the arrester layout data corresponding to the zinc oxide arrester group; and performing multi-level fault coupling optimization on the arrester fault first graph according to the arrester topology distribution model to obtain an arrester fault second graph. The application can solve the technical problem of low accuracy of the abnormal state monitoring of zinc oxide arresters in the prior art, and achieve the technical effect of improving the accuracy of abnormal state monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of measuring power transformation, and in particular to a method and system for monitoring abnormal states of zinc oxide arresters in distribution lines. BACKGROUND

[0002] Arresters are mainly used to introduce overvoltage into the ground network in the case of lightning surges, power grid operating overvoltage or transient interference, so as to prevent the insulation of transformers, switching devices and lines from being punctured, thereby ensuring the stable operation of the system. However, with the development of urban power grids towards high voltage, complexity and distribution, the operating environment and electrical load of arresters are constantly diversifying, and their state degradation characteristics show concealment, coupling and dynamics, so that the traditional methods of monitoring and predicting arresters have been difficult to meet the needs of safe operation of power systems.

[0003] At present, the existing state monitoring of arresters mainly relies on periodic manual inspection, offline insulation testing or simple current and voltage parameter acquisition, and these methods have obvious limitations. On the one hand, manual inspection has the problems of long cycle and poor real-time performance, and cannot timely capture the operating state of arresters at the moment of lightning strike or in the case of severe environmental changes; on the other hand, the traditional parameter acquisition method can only monitor a single electrical indicator, such as leakage current or residual voltage, and lacks comprehensive analysis capability for multi-dimensional environmental factors, fault evolution trend and neighborhood influence.

[0004] In summary, in the prior art, there is a technical problem that the operating state of arresters cannot be accurately perceived in real time due to the fact that the monitoring means is too single, the data dimension is limited, and there is a lack of comprehensive analysis capability for the coupling relationship between environmental disturbance, fault evolution and arrester groups, potential faults are difficult to be identified and intervened early, and further affect the safe and stable operation of power systems, the scientificity and foresight of equipment full life cycle management and operation and maintenance decision-making. SUMMARY

[0005] The purpose of the present application is to provide a method and system for monitoring abnormal states of zinc oxide arresters in distribution lines, so as to solve the technical problem in the prior art that the operating state of arresters cannot be accurately perceived in real time due to the fact that the monitoring means is too single, the data dimension is limited, and there is a lack of comprehensive analysis capability for the coupling relationship between environmental disturbance, fault evolution and arrester groups, potential faults are difficult to be identified and intervened early, and further affect the safe and stable operation of power systems, the scientificity and foresight of equipment full life cycle management and operation and maintenance decision-making.

[0006] In view of the above problems, the present application provides a method and system for monitoring abnormal states of zinc oxide arresters in distribution lines.

[0007] In a first aspect, the application provides an abnormal state monitoring method of a zinc oxide arrester in a power distribution line, which is realized by an abnormal state monitoring system of the zinc oxide arrester in the power distribution line, and includes the following steps: monitoring a group of zinc oxide arresters in the power distribution line in real time to obtain an arrester monitoring data block; performing multi-dimensional conduction accompanying fault prediction under balanced counteraction optimization on the group of zinc oxide arresters according to the arrester monitoring data block to obtain a conduction accompanying fault atlas; performing multi-dimensional non-conduction accompanying fault prediction under balanced counteraction learning on the group of zinc oxide arresters according to the arrester monitoring data block to obtain a non-conduction accompanying fault atlas; constructing an arrester topology distribution model according to arrester layout data corresponding to the group of zinc oxide arresters; performing neighborhood action intervention compensation on the conduction accompanying fault atlas and the non-conduction accompanying fault atlas according to the arrester topology distribution model to construct an arrester fault first atlas; and performing multi-level fault coupling optimization on the arrester fault first atlas according to the arrester topology distribution model to obtain an arrester fault second atlas.

[0008] Preferably, the abnormal state monitoring method of the zinc oxide arrester in the power distribution line further includes the following steps: extracting an a-th arrester monitoring patch block corresponding to an a-th arrester according to the arrester monitoring data block, and interconnecting the a-th arrester with zinc oxide arresters of the same type according to the a-th arrester to obtain an a-th arrester cluster, where a is a positive integer; performing conduction accompanying fault history retrieval on the a-th arrester cluster to obtain an a-th conduction accompanying fault history set; performing supervised training on Q learners according to the a-th conduction accompanying fault history set to obtain Q conduction accompanying fault prediction models, where Q is a positive integer greater than 1; performing balanced counteraction optimization aggregation on the Q conduction accompanying fault prediction models according to a predetermined balanced evaluation value to obtain an a-th conduction accompanying fault prediction node; inputting the a-th arrester monitoring patch block into the a-th conduction accompanying fault prediction node to obtain an a-th conduction accompanying fault prediction result, and adding the a-th conduction accompanying fault prediction result to the conduction accompanying fault atlas.

[0009] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: classifying the a-th conduction accompanied fault history set according to a plurality of conduction accompanied fault indicators, obtaining a plurality of conduction accompanied fault partitions; sample balance evaluation is performed on the plurality of conduction accompanied fault partitions to obtain a sample balance evaluation value; difference identification is performed on the sample balance evaluation value according to the predetermined balance evaluation value to obtain a balance evaluation difference feature; the balance evaluation difference feature is traced back to the plurality of conduction accompanied fault partitions to determine a balance difference trace factor; the plurality of conduction accompanied fault partitions are subjected to counter sample injection according to the balance difference trace factor to obtain an a-th conduction accompanied fault optimization set; the Q conduction accompanied fault prediction models are tested according to the a-th conduction accompanied fault optimization set to obtain Q conduction accompanied fault loss sets; and the Q conduction accompanied fault prediction models are optimized and aggregated for training to generate the a-th conduction accompanied fault prediction node.

[0010] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: identifying the neighborhood of each zinc oxide arrester in the zinc oxide arrester group according to the arrester topology distribution model to obtain a plurality of arrester neighborhoods; performing action interference compensation on the conduction fault spectrum according to the plurality of arrester neighborhoods to obtain a conduction fault optimization spectrum; performing action interference compensation on the non-conduction accompanied fault spectrum according to the plurality of arrester neighborhoods to obtain a non-conduction fault optimization spectrum; performing conflict detection according to the conduction fault optimization spectrum and the non-conduction fault optimization spectrum to obtain a fault conflict feature distribution; and fusing the conduction fault optimization spectrum and the non-conduction fault optimization spectrum according to the fault conflict feature distribution to obtain the arrester fault first spectrum.

[0011] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: extracting an a-th arrester neighborhood corresponding to an a-th arrester according to the plurality of arrester neighborhoods, and collecting real-time action parameters according to the a-th arrester neighborhood to obtain a-th neighborhood action data; performing abnormality identification according to the a-th neighborhood action data to obtain a-th neighborhood action abnormality features; activating a neighborhood interference trace architecture, the neighborhood interference trace architecture comprising neighborhood interference features, interference-induced intermediate features, and interference-induced basic features; performing action interference tracing on the a-th arrester according to the neighborhood interference trace architecture based on the a-th neighborhood action abnormality features to determine a a-th neighborhood action interference path; correlating and compensating an a-th conduction accompanied fault prediction result according to the a-th neighborhood action interference path to obtain an a-th conduction accompanied fault optimization result, and adding the a-th conduction accompanied fault optimization result to the conduction fault optimization spectrum.

[0012] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: monitoring the power distribution line in real time to obtain a line monitoring data set; performing fault prediction on the power distribution line according to the line monitoring data set to obtain a line fault distribution; coupling fault deduction is performed on the arrester fault first graph according to the line fault distribution based on the arrester topology distribution model to obtain a first coupled fault distribution; environment coupling fault deduction is performed on the arrester fault first graph according to the arrester topology distribution model to obtain a second coupled fault distribution; and the arrester fault second graph is generated by optimizing the arrester fault first graph according to the first coupled fault distribution and the second coupled fault distribution.

[0013] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: performing environment prediction according to the group of zinc oxide arresters to obtain an environment sequence of each arrester; performing environment anomaly attention reinforcement according to the environment sequence of each arrester to obtain an environment vector of each arrester; performing trend deduction on the arrester fault first graph according to the environment vector of each arrester based on the arrester topology distribution model to obtain a trend deduction result of each fault; and performing environment factor tracing on the trend deduction result of each fault according to the environment vector of each arrester to obtain the second coupled fault distribution.

[0014] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: monitoring the group of zinc oxide arresters in real time to obtain an arrester data set; performing data cleaning and blockchain packaging on the arrester data set to generate the arrester monitoring data block.

[0015] Preferably, the abnormal state monitoring method of zinc oxide arresters in the power distribution line further comprises: generating an arrester early warning instruction according to the arrester fault second graph.

[0016] In a second aspect, the application also provides an abnormal state monitoring system of zinc oxide arresters in a power distribution line, which is used to execute the abnormal state monitoring method of zinc oxide arresters in a power distribution line as described in the first aspect, and comprises: an arrester monitoring data block obtaining module, which is used to monitor zinc oxide arresters in a power distribution line in real time and obtain arrester monitoring data blocks; a conduction accompanying fault atlas obtaining module, which is used to perform multi-dimensional conduction accompanying fault prediction of the zinc oxide arrester group under balanced confrontation optimization according to the arrester monitoring data blocks and obtain a conduction accompanying fault atlas; a non-conduction accompanying fault atlas obtaining module, which is used to perform multi-dimensional non-conduction accompanying fault prediction of the zinc oxide arrester group under balanced confrontation learning according to the arrester monitoring data blocks and obtain a non-conduction accompanying fault atlas; an arrester topology distribution model constructing module, which is used to construct an arrester topology distribution model according to arrester layout data corresponding to the zinc oxide arrester group; an arrester fault first atlas constructing module, which is used to perform neighborhood action interference compensation on the conduction accompanying fault atlas and the non-conduction accompanying fault atlas according to the arrester topology distribution model and construct an arrester fault first atlas; and an arrester fault second atlas obtaining module, which is used to perform multi-level fault coupling optimization on the arrester fault first atlas according to the arrester topology distribution model and obtain an arrester fault second atlas.

[0017] The technical solutions provided in the application have at least the following technical effects or advantages: by implementing the construction of an arrester intelligent state monitoring system facing multi-source perception and dynamic correlation analysis, the technical goal of having multi-dimensional recognition, trend prediction and active early warning capability for the running state, environmental disturbance and group interaction effect is achieved, the full-time domain, full-element and global monitoring and intelligent diagnosis of the arrester state are realized, and the technical effects of significantly improving the protection capability, fault response speed and operation and maintenance level of the power system are achieved.

[0018] The above description is only a summary of the technical solutions of the application. In order to enable one of ordinary skill in the art to better understand the technical means of the application and implement it according to the content of the description, and in order to enable the above and other purposes, features and advantages of the application to be more apparent and easy to understand, the specific implementation manner of the application is described below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the application, nor is it used to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can obtain other drawings according to the provided drawings without any creative effort.

[0020] Figure 1 FIG. 1 is a flowchart of an abnormal state monitoring method of a zinc oxide arrester in a power distribution line according to an embodiment of the present application.

[0021] Figure 2 FIG. 2 is a structural diagram of an abnormal state monitoring system of a zinc oxide arrester in a power distribution line according to an embodiment of the present application.

[0022] FIG. 1 is a flowchart of an abnormal state monitoring method of a zinc oxide arrester in a power distribution line according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The present application provides an abnormal state monitoring method and system of a zinc oxide arrester in a power distribution line, which solves the technical problem in the prior art that the operation state of the arrester cannot be perceived in real time and accurately, potential faults are difficult to be identified and intervened early, and the safety and stability of the power system, the equipment life cycle management, and the scientificity and foresight of the operation and maintenance decision are further affected due to the single monitoring means, the limited data dimension, and the lack of comprehensive analysis capability of the coupling relationship between the environmental disturbance, the fault evolution, and the arrester group. The technical goal of constructing an arrester intelligent state monitoring system facing multi-source perception and dynamic correlation analysis is achieved, which has the multi-dimensional identification, trend prediction, and active early warning capability of the operation state, the environmental disturbance, and the group interaction effect, and the technical effect of realizing the full-time, full-element, and global monitoring and intelligent diagnosis of the arrester state is achieved, which significantly improves the protection capability, the fault response speed, and the operation and maintenance level of the power system.

[0024] The technical solutions in the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application. In addition, it should be noted that, for convenience of description, only parts related to the present application are shown in the drawings, rather than all.

[0025] Embodiment one, please refer to the accompanying Figure 1 The present application provides an abnormal state monitoring method of a zinc oxide arrester in a power distribution line, which is applied to an abnormal state monitoring system of a zinc oxide arrester in a power distribution line, and specifically includes the following steps:

[0026] S1: Real-time monitoring of a zinc oxide arrester group in a power distribution line to obtain an arrester monitoring data block.

[0027] Further, the application also includes: real-time monitoring of the zinc oxide arrester group to obtain an arrester data set; data cleaning and blockchain packaging of the arrester data set to generate the arrester monitoring data block.

[0028] Specifically, real-time monitoring of the zinc oxide arrester group means continuously obtaining the running state information of each zinc oxide arrester through sensors or monitoring devices installed on the distribution line, including voltage, current, temperature, leakage current and other parameters, thereby forming complete arrester group running data. Zinc oxide arrester is an electrical component used to protect distribution lines and equipment from lightning strikes or overvoltage damage. By real-time monitoring, continuous and instant data are collected to reflect the dynamic state of the arrester. The arrester data set refers to organizing the real-time collected data by arrester or time sequence to form a data collection that is convenient for processing and analysis.

[0029] Then, data cleaning and blockchain packaging of the arrester data set means preprocessing the collected raw data to remove outliers, duplicates and noise, and simultaneously interpolating missing data to maintain data consistency and reliability, thereby improving the accuracy of subsequent analysis. Blockchain packaging means encrypting and storing the cleaned data in a distributed manner according to blockchain technology to ensure data tamper resistance and traceability, forming a secure and reliable data block. The arrester monitoring data block refers to a structured data unit that can be directly used for fault prediction and optimization analysis after cleaning and blockchain processing.

[0030] S2: Multi-dimensional conduction accompanying fault prediction of the zinc oxide arrester group under balanced counter-optimization according to the arrester monitoring data block to obtain a conduction accompanying fault atlas.

[0031] Further, the application also includes: extracting an a-th arrester monitoring slice block corresponding to an a-th arrester according to the arrester monitoring data block, and interconnecting the a-th arrester with zinc oxide arresters of the same model according to the a-th arrester to obtain an a-th arrester cluster, a being a positive integer; performing conduction accompanying fault history retrieval on the a-th arrester cluster to obtain an a-th conduction accompanying fault history set; performing supervised training on Q learning machines according to the a-th conduction accompanying fault history set to obtain Q conduction accompanying fault prediction models, Q being a positive integer greater than 1; performing balanced counter-optimization aggregation on the Q conduction accompanying fault prediction models according to a predetermined balanced evaluation value to obtain an a-th conduction accompanying fault prediction node; inputting the a-th arrester monitoring slice block into the a-th conduction accompanying fault prediction node to obtain an a-th conduction accompanying fault prediction result, and adding the a-th conduction accompanying fault prediction result to the conduction accompanying fault atlas.

[0032] Further, the application also includes: classifying the a-th conduction accompanying fault history set according to a plurality of conduction accompanying fault indicators to obtain a plurality of conduction accompanying fault partitions; performing sample balance evaluation on the plurality of conduction accompanying fault partitions to obtain a sample balance evaluation value; performing difference identification on the sample balance evaluation value according to the predetermined balance evaluation value to obtain a balance evaluation difference feature; performing trace analysis on the balance evaluation difference feature according to the plurality of conduction accompanying fault partitions to determine a balance difference trace factor; performing counter sample injection on the plurality of conduction accompanying fault partitions according to the balance difference trace factor to obtain an a-th conduction accompanying fault optimization set; testing the Q conduction accompanying fault prediction models according to the a-th conduction accompanying fault optimization set to obtain Q conduction accompanying fault loss sets; and performing optimized aggregation training on the Q conduction accompanying fault prediction models according to the Q conduction accompanying fault loss sets to generate the a-th conduction accompanying fault prediction node.

[0033] Specifically, according to the lightning arrester monitoring data block, the a-th lightning arrester monitoring piece block corresponding to the a-th lightning arrester is extracted, that is, the relevant monitoring data for the a-th lightning arrester is screened out from the entire lightning arrester monitoring data block, referred to as the a-th lightning arrester monitoring piece block, including voltage, leakage current, temperature and other state information. According to the a-th lightning arrester, the same type of zinc oxide lightning arrester is interconnected to obtain the a-th lightning arrester cluster, that is, in order to enhance the prediction accuracy, the data of other lightning arresters of the same type as the a-th lightning arrester are interconnected to form a cluster, referred to as the a-th lightning arrester cluster. a is a positive integer, used to identify the a-th lightning arrester.

[0034] Then, according to the a-th lightning arrester cluster, the conduction accompanying fault history is retrieved to obtain the a-th conduction accompanying fault history set, which means that the past conduction state related fault records are searched in the a-th lightning arrester cluster, including the abnormal action data of each lightning arrester under different time or different environmental conditions, forming the a-th conduction accompanying fault history set, providing a historical basis for subsequent prediction.

[0035] Next, the Q learning machines are supervised trained according to the a-th conduction accompanying fault history set to obtain Q conduction accompanying fault prediction models, Q is a positive integer greater than 1, which means that the a-th conduction accompanying fault history set is input into a plurality of machine learning models for training, and the mapping relationship between the historical data and the fault occurrence is learned through supervised learning, thereby generating Q prediction models, each model can independently predict future conduction accompanying faults, and the value of Q is greater than 1, which can increase the diversity and prediction robustness of the model.

[0036] Further, the a-th conduction-companion fault history set is classified according to a plurality of conduction-companion fault indicators, and a plurality of conduction-companion fault partitions are obtained, which means that each type of fault event recorded in the a-th conduction-companion fault history set is grouped according to different indicators, such as overcurrent amplitude, action delay time, environmental temperature, etc., and each type of fault event forms an independent partition, thereby facilitating subsequent targeted analysis and processing of different types of faults.

[0037] Then, sample balance evaluation is performed on the plurality of conduction-companion fault partitions, and a sample balance evaluation value is obtained, which means that the number of samples in each partition is counted, and the balance of the samples in each partition is evaluated according to a predetermined standard. For example, a small number of samples in some partitions may cause model training bias, and the sample distribution of each partition is reflected by calculating the balance evaluation value.

[0038] Next, a difference identification is performed on the sample balance evaluation value according to a predetermined balance evaluation value, and a balance evaluation difference feature is obtained, which means that the actual sample balance evaluation value is compared and analyzed with the ideal balance evaluation value, and the partitions with obvious sample distribution differences are identified, thereby generating a balance evaluation difference feature reflecting the size and direction of the bias.

[0039] Then, the balance difference traceability factor is determined by performing traceability analysis on the balance evaluation difference feature according to the plurality of conduction-companion fault partitions, which means that the potential causes of the sample distribution difference are analyzed, such as historical data missing, environmental abnormalities or specific arrester characteristics, thereby determining the traceability factor of the balance difference of each partition, which provides a basis for subsequent adversarial sample injection.

[0040] Subsequently, adversarial samples are injected into the plurality of conduction-companion fault partitions according to the balance difference traceability factor, and an a-th conduction-companion fault optimization set is obtained, which means that additional synthetic samples are generated or existing sample features are adjusted in partitions with insufficient samples or obvious bias, so that the overall sample distribution is more balanced, thereby forming an optimized fault history set.

[0041] Then, the Q conduction-companion fault prediction models are tested according to the a-th conduction-companion fault optimization set, and Q conduction-companion fault loss sets are obtained, which means that the optimized sample set is input into each prediction model, the error or loss of the model in each type of fault prediction is calculated, and a loss set corresponding to each model is formed to evaluate the prediction effect of the model.

[0042] Finally, the Q conduction-companion fault prediction models are optimized and aggregated for training according to the Q conduction-companion fault loss sets, and an a-th conduction-companion fault prediction node is generated, which means that the loss set is used to adjust the weights of each model and jointly train the models, the overall prediction ability of the models is optimized through the balance and adversarial strategy, and finally an aggregated prediction node is formed, which can more accurately predict the conduction-companion fault of the a-th arrester.

[0043] Finally, input the a arrester monitoring piece into the a conduction accompanying fault prediction node, obtain the a conduction accompanying fault prediction result, and add the a conduction accompanying fault prediction result to the conduction accompanying fault atlas, that is, input the a arrester current monitoring data into the aggregated prediction node, obtain the fault prediction result of the arrester in the conduction state, and then update the prediction result to the conduction accompanying fault atlas of the entire system, so as to form a global fault distribution view.

[0044] S3: According to the arrester monitoring data block, the multi-dimensional non-conduction accompanying fault prediction of the zinc oxide arrester group under balanced adversarial learning is obtained, and a non-conduction accompanying fault atlas is obtained.

[0045] Specifically, according to the arrester monitoring data block, the multi-dimensional non-conduction accompanying fault prediction of the zinc oxide arrester group under balanced adversarial learning is obtained, and a non-conduction accompanying fault atlas is obtained. Non-conduction state refers to the running state of the arrester without current conduction, that is, the overvoltage protection action is not triggered, while the accompanying fault refers to the abnormal condition that may occur during the non-conduction of the arrester due to environmental factors, voltage fluctuations or device aging, such as local insulation deterioration or leakage current anomaly. Balanced adversarial learning is a machine learning strategy that balances learning on various non-conduction abnormal conditions by constructing adversarial samples or adjusting the weights of training samples, avoiding bias towards a certain type of fault, thereby improving the accuracy and robustness of prediction. Multi-dimensional means that the prediction not only considers a single index, but also analyzes multiple dimensions of information such as voltage, temperature, leakage current and environmental humidity.

[0046] Then, obtaining the non-conduction accompanying fault atlas means presenting the prediction results in the form of atlas, each arrester corresponding to a group of non-conduction fault probabilities or abnormal feature indexes, forming a visual or structured data representation for further analysis, fault correlation or early warning decision. Multi-dimensional information can display the potential risk distribution of each arrester in different environmental conditions under non-conduction state, providing basis for maintenance and prevention.

[0047] S4: According to the arrester layout data corresponding to the zinc oxide arrester group, a arrester topology distribution model is constructed.

[0048] Specifically, according to the arrester layout data corresponding to the zinc oxide arrester group, a topological distribution model of the arrester is constructed, which refers to using the installation position, connection mode, electrical parameters and line direction of each zinc oxide arrester on the distribution line and other layout information to establish a model that can reflect the spatial structure and electrical relationship of the arrester group. The layout data includes the specific installation point of each arrester in the distribution line, the electrical connection mode with adjacent arresters or equipment, and the type and capacity parameters of the arrester. The topological distribution model refers to the graphical or mathematical structure constructed according to the layout data, which can describe the adjacency relationship, path connectivity and network level between arresters, facilitating the analysis of the influence of each arrester on fault propagation or interference.

[0049] Then, the purpose of constructing the topological distribution model of the arrester is to provide a structure basis for subsequent fault prediction and optimization. Through the topological model, the neighborhood of each arrester can be identified, that is, which arresters may influence each other, and when an abnormality occurs in a certain arrester, along which paths the fault may propagate, thereby realizing neighborhood interference compensation and multi-level coupling analysis.

[0050] S5: According to the arrester topological distribution model, the conduction accompanied fault atlas and the non-conduction accompanied fault atlas are compensated for neighborhood action interference, and a first arrester fault atlas is constructed.

[0051] Further, the present application also includes: according to the arrester topological distribution model, the neighborhood of each zinc oxide arrester in the zinc oxide arrester group is identified, and a plurality of arrester neighborhoods are obtained; according to the plurality of arrester neighborhoods, the conduction accompanied fault atlas is compensated for action interference, and a conduction fault optimization atlas is obtained; according to the plurality of arrester neighborhoods, the non-conduction accompanied fault atlas is compensated for action interference, and a non-conduction fault optimization atlas is obtained; according to the conduction fault optimization atlas and the non-conduction fault optimization atlas, conflict detection is performed, and a fault conflict feature distribution is obtained; according to the fault conflict feature distribution, the conduction fault optimization atlas and the non-conduction fault optimization atlas are fused, and the first arrester fault atlas is obtained.

[0052] Further, the application also includes: according to the plurality of arrester neighborhoods, extracting the a-th arrester neighborhood corresponding to the a-th arrester, and collecting real-time action parameters according to the a-th arrester neighborhood to obtain a-th neighborhood action data; identifying abnormalities according to the a-th neighborhood action data to obtain a-th neighborhood action abnormal features; activating a neighborhood interference tracing architecture, which includes neighborhood interference feature-interference induced intermediate features and interference induced basic features; based on the a-th neighborhood action abnormal features, performing action interference tracing on the a-th arrester according to the neighborhood interference tracing architecture to determine an a-th neighborhood action interference path; associating and compensating the a-th conduction accompanying fault prediction result according to the a-th neighborhood action interference path to obtain an a-th conduction accompanying fault optimization result, and adding the a-th conduction accompanying fault optimization result to the conduction fault optimization atlas.

[0053] Specifically, according to the arrester topology distribution model, the neighborhood of each zinc oxide arrester in the zinc oxide arrester group is identified to obtain a plurality of arrester neighborhoods, which means that the electrical and spatial relationship between each arrester and the surrounding arrester is analyzed by using the arrester topology distribution model constructed in the foregoing to determine the neighborhood range of each arrester, that is, the arrester set that may affect each other is identified, which is called an arrester neighborhood, and each neighborhood contains a number of neighbor nodes directly related to the target arrester.

[0054] Further, according to the plurality of arrester neighborhoods, the a-th arrester neighborhood corresponding to the a-th arrester is extracted, and real-time action parameters are collected according to the a-th arrester neighborhood to obtain a-th neighborhood action data, which means that in the neighborhood set of the entire arrester group, the neighbor set corresponding to the a-th arrester is found, and each arrester in the neighborhood is monitored in real time to collect action-related parameters, such as overvoltage trigger time, current amplitude, leakage current change or temperature fluctuation. The collected data is called a-th neighborhood action data, which is used to reflect the dynamic behavior of the arrester in the neighborhood.

[0055] Then, according to the a-th neighborhood action data, abnormality is identified to obtain a-th neighborhood action abnormal features, which means that the a-th neighborhood action data is analyzed by using an algorithm to identify abnormal conditions deviating from the normal operation range or historical patterns, such as a certain arrester action delay exceeding a preset threshold or abnormal increase of leakage current. The abnormal index forms the a-th neighborhood action abnormal features, which are used to represent potential faults or disturbances in the neighborhood.

[0056] Then, the neighborhood interference tracing architecture is activated, which includes neighborhood interference features, interference-induced intermediate features, and interference-induced basic features, refers to starting a system for analyzing neighborhood interference sources and propagation paths. The architecture maps abnormal features to intermediate induced features and further correlates them to basic induced features, establishing a hierarchical relationship from abnormal behavior to interference causes for tracing interference sources.

[0057] Then, based on the a-th neighborhood action abnormal feature, the a-th arrester is subjected to action interference tracing according to the neighborhood interference tracing architecture, and the a-th neighborhood action interference path is determined, which means that the a-th neighborhood action abnormality is mapped to the neighborhood interference factors that may cause the abnormality using the tracing architecture, thereby forming the a-th neighborhood action interference path, i.e., determining which neighbor arrester actions may have affected the a-th arrester behavior.

[0058] Finally, the a-th conduction concomitant fault prediction result is associated and compensated according to the a-th neighborhood action interference path to obtain the a-th conduction concomitant fault optimization result, and the a-th conduction concomitant fault optimization result is added to the conduction fault optimization atlas, which means that the original prediction of the a-th arrester is corrected, such as increasing or decreasing the prediction probability, so that it is closer to the actual risk, and the optimized result is updated to the global conduction fault optimization atlas, realizing the accuracy of the overall prediction.

[0059] Further, the non-conduction concomitant fault atlas is subjected to action interference compensation according to the neighborhood of multiple arresters to obtain a non-conduction fault optimization atlas, which means that the neighborhood interference correction is also performed on the concomitant fault atlas in the non-conduction state, and the prediction value or abnormal index is adjusted so that the non-conduction fault optimization atlas can more accurately reflect the potential risks of the arrester in the non-conduction state.

[0060] Then, conflict detection is performed according to the conduction fault optimization atlas and the non-conduction fault optimization atlas to obtain a fault conflict feature distribution, which means that the prediction conflicts or inconsistencies that may exist in the conduction and non-conduction fault optimization atlases are analyzed, such as the fault indicators of the same arrester in the conduction and non-conduction states contradicting each other. The fault conflict feature distribution is formed by detecting the conflict for subsequent fusion processing.

[0061] Finally, the conduction fault optimization atlas and the non-conduction fault optimization atlas are fused according to the fault conflict feature distribution to obtain an arrester fault first atlas, which means that the optimization atlases in the conduction and non-conduction states are combined, the conflict features are weighted or corrected, and a unified arrester fault first atlas is generated, which can comprehensively reflect the potential fault distribution of each arrester in different states.

[0062] S6: performing multi-level fault coupling optimization on the arrester fault first graph according to the arrester topology distribution model to obtain an arrester fault second graph.

[0063] Further, the application further comprises: performing real-time monitoring on the power distribution line to obtain a line monitoring data set; performing fault prediction on the power distribution line according to the line monitoring data set to obtain a line fault distribution; performing coupling fault deduction on the arrester fault first graph according to the line fault distribution based on the arrester topology distribution model to obtain a first coupling fault distribution; performing environment coupling fault deduction on the arrester fault first graph according to the arrester topology distribution model to obtain a second coupling fault distribution; and optimizing the arrester fault first graph according to the first coupling fault distribution and the second coupling fault distribution to generate the arrester fault second graph.

[0064] Further, the application further comprises: performing environment prediction according to the zinc oxide arrester group to obtain an environment sequence of each arrester; performing environment anomaly attention reinforcement according to the environment sequence of each arrester to obtain an environment vector of each arrester; performing trend deduction on the arrester fault first graph according to the environment vector of each arrester based on the arrester topology distribution model to obtain a trend deduction result of each fault; and performing environment factor tracing on the trend deduction result of each fault according to the environment vector of each arrester to obtain the second coupling fault distribution.

[0065] Further, the application further comprises: generating an arrester early warning instruction according to the arrester fault second graph.

[0066] Specifically, the real-time monitoring of the power distribution line to obtain a line monitoring data set means that various sensors installed on the power distribution line, such as voltage, current, temperature, humidity and partial discharge sensors, continuously monitor the line operation state, real-time collect various operation parameters, form a structured line monitoring data set, and record the multi-dimensional operation information of the power distribution line at different time points, providing a basis for subsequent fault analysis.

[0067] Then, the fault prediction of the power distribution line according to the line monitoring data set to obtain a line fault distribution means that the collected line monitoring data is analyzed by an algorithm model to analyze possible fault types and positions, such as overload, short circuit or insulation deterioration, and generate corresponding fault probability or risk index, and the prediction result forms a line fault distribution in space, which can display the potential fault risk level of each line section or node.

[0068] Then, based on the arrester topology distribution model, the arrester fault first graph is coupled with the fault deduction according to the line fault distribution, and a first coupled fault distribution is obtained. It is referred to as mapping the influence that the line fault may cause to each arrester by combining the topology relationship of the arrester in the line, considering the fault propagation path and neighborhood correlation, so as to obtain the first coupled fault distribution, that is, the predicted distribution of each arrester affected by the line fault in the conducting or non-conducting state.

[0069] Further, the environment of the zinc oxide arrester group is predicted to obtain an environment sequence of each arrester. It is referred to as dynamically predicting the operating environment of the zinc oxide arrester installed at different positions, including temperature, humidity, air pressure, salt density, lightning frequency, pollution level and wind speed and other environmental parameters. The future change trend is calculated through a time series model, so as to generate a continuous environment data sequence for each arrester, which is used to represent the evolution law of the external environment over time.

[0070] Then, the environment of each arrester is abnormal and the attention is strengthened to obtain an environment vector of each arrester. It is referred to as using the attention mechanism in deep learning to extract and weight the abnormal part in the environment sequence that has a greater impact on the operation state of the arrester, such as high humidity, lightning density or sharp increase in pollution and other factors. The environment vector generated is a high-dimensional representation of the key influence features in the environment sequence, which is used for subsequent fault trend analysis and risk assessment.

[0071] Then, based on the arrester topology distribution model, the arrester fault first graph is coupled with the fault deduction according to the environment vector of each arrester to obtain a fault trend deduction result of each arrester. It is referred to as combining the extracted environment vector with the topology structure of the arrester to simulate the long-term influence of environmental factors on the overall fault behavior of the arrester group, so as to dynamically evolve and predict the arrester fault first graph, obtain the fault trend that each arrester may appear, such as rising of conduction probability, acceleration of insulation degradation or delay of non-conducting response, and form a trend deduction result.

[0072] Subsequently, the environment factor of each arrester environment vector is traced to obtain a second coupled fault distribution. It is referred to as identifying the specific environmental factors that cause different fault trends by reverse analysis of the trend deduction result, and associating with the corresponding arrester node to form a second coupled fault distribution containing the environmental influence path and the action mechanism, so as to describe the mapping relationship between the external environment and the arrester fault behavior.

[0073] Finally, according to the first coupling fault distribution and the second coupling fault distribution, the lightning arrester fault first graph is optimized to generate a lightning arrester fault second graph, which means that the line fault coupling distribution and the environmental coupling distribution are comprehensively analyzed, the lightning arrester fault first graph is adjusted and optimized, for example, the fault probability or weight is corrected, and finally the lightning arrester fault second graph is generated, which can comprehensively reflect the potential risk distribution of the lightning arrester under the influence of line faults and environment in different states.

[0074] Further, according to the lightning arrester fault second graph, a lightning arrester early warning instruction is generated, which means that after obtaining the lightning arrester fault second graph optimized through multi-dimensional fault analysis and coupling deduction, the fault distribution, risk level, trend evolution and correlation contained in the lightning arrester fault second graph are analyzed, thereby automatically generating an early warning instruction for operation and maintenance management and safety protection. The lightning arrester fault second graph means a comprehensive risk graph that integrates conduction faults, non-conduction faults, environmental influences and line coupling, which not only reflects the current health status of each lightning arrester, but also embodies the development trend and possible chain reaction of potential faults. The early warning instruction is an executable operation information, including risk level classification, response time suggestion, maintenance priority sorting and necessary control strategy, such as real-time monitoring encryption of high-risk lightning arrester, advance arrangement of maintenance plan, or isolation of operation node to prevent fault propagation when necessary.

[0075] In summary, the abnormal state monitoring method of the zinc oxide lightning arrester in the power distribution line provided in the present application has the following technical effects: by realizing the construction of the lightning arrester intelligent state monitoring system facing multi-source sensing and dynamic correlation analysis, the technical goal of multi-dimensional identification, trend prediction and active early warning capability of the running state, environmental disturbance and group interaction effect is achieved, the full-time domain, full-element and global monitoring and intelligent diagnosis of the lightning arrester state are realized, and the technical effects of significantly improving the power system protection capability, fault response speed and operation and maintenance level are achieved.

[0076] Embodiment two, based on the same inventive concept as the abnormal state monitoring method of the zinc oxide lightning arrester in the power distribution line in the foregoing embodiments, the present application also provides an abnormal state monitoring system of the zinc oxide lightning arrester in the power distribution line, please refer to the attached Figure 2The system comprises: a lightning arrester monitoring data block obtaining module 1, which is used for monitoring a zinc oxide lightning arrester group of a power distribution line in real time and obtaining lightning arrester monitoring data blocks; a conduction accompanying fault graph obtaining module 2, which is used for performing multi-dimensional conduction accompanying fault prediction of the zinc oxide lightning arrester group under balanced confrontation optimization according to the lightning arrester monitoring data blocks and obtaining a conduction accompanying fault graph; a non-conduction accompanying fault graph obtaining module 3, which is used for performing multi-dimensional non-conduction accompanying fault prediction of the zinc oxide lightning arrester group under balanced confrontation learning according to the lightning arrester monitoring data blocks and obtaining a non-conduction accompanying fault graph; a lightning arrester topology distribution model constructing module 4, which is used for constructing a lightning arrester topology distribution model according to lightning arrester layout data corresponding to the zinc oxide lightning arrester group; a lightning arrester fault first graph constructing module 5, which is used for performing neighborhood action interference compensation on the conduction accompanying fault graph and the non-conduction accompanying fault graph according to the lightning arrester topology distribution model and constructing a lightning arrester fault first graph; and a lightning arrester fault second graph obtaining module 6, which is used for performing multi-level fault coupling optimization on the lightning arrester fault first graph according to the lightning arrester topology distribution model and obtaining a lightning arrester fault second graph.

[0077] Further, the abnormal state monitoring system of the zinc oxide lightning arrester in the power distribution line is also used for: extracting an a-th lightning arrester monitoring piece block corresponding to an a-th lightning arrester according to the lightning arrester monitoring data blocks, and performing interconnection of lightning arresters of the same model according to the a-th lightning arrester to obtain an a-th lightning arrester cluster, a being a positive integer; performing conduction accompanying fault history retrieval according to the a-th lightning arrester cluster to obtain an a-th conduction accompanying fault history set; performing supervised training of Q learning machines according to the a-th conduction accompanying fault history set to obtain Q conduction accompanying fault prediction models, Q being a positive integer greater than 1; performing balanced confrontation optimization aggregation of the Q conduction accompanying fault prediction models according to a predetermined balanced evaluation value to obtain an a-th conduction accompanying fault prediction node; inputting the a-th lightning arrester monitoring piece block into the a-th conduction accompanying fault prediction node to obtain an a-th conduction accompanying fault prediction result, and adding the a-th conduction accompanying fault prediction result to the conduction accompanying fault graph.

[0078] Further, the abnormal state monitoring system of zinc oxide surge arresters in power distribution lines is also used for: classifying the a-th conduction accompanied fault history set according to a plurality of conduction accompanied fault indicators, obtaining a plurality of conduction accompanied fault partitions; sample balance evaluation is performed on the plurality of conduction accompanied fault partitions, and a sample balance evaluation value is obtained; difference identification is performed on the sample balance evaluation value according to the predetermined balance evaluation value, and a balance evaluation difference feature is obtained; the balance evaluation difference feature is traced and analyzed according to the plurality of conduction accompanied fault partitions, and a balance difference trace factor is determined; the plurality of conduction accompanied fault partitions are subjected to counter sample injection according to the balance difference trace factor, and an a-th conduction accompanied fault optimization set is obtained; the Q conduction accompanied fault prediction models are tested according to the a-th conduction accompanied fault optimization set, and Q conduction accompanied fault loss sets are obtained; the Q conduction accompanied fault prediction models are optimized and aggregated for training according to the Q conduction accompanied fault loss sets, and the a-th conduction accompanied fault prediction node is generated.

[0079] Further, the abnormal state monitoring system of zinc oxide surge arresters in power distribution lines is also used for: according to the surge arrester topology distribution model, each zinc oxide surge arrester in the zinc oxide surge arrester group is subjected to neighborhood identification, and a plurality of surge arrester neighborhoods are obtained; the conduction accompanied fault atlas is subjected to action interference compensation according to the plurality of surge arrester neighborhoods, and a conduction fault optimization atlas is obtained; the non-conduction accompanied fault atlas is subjected to action interference compensation according to the plurality of surge arrester neighborhoods, and a non-conduction fault optimization atlas is obtained; conflict detection is performed according to the conduction fault optimization atlas and the non-conduction fault optimization atlas, and a fault conflict feature distribution is obtained; the conduction fault optimization atlas and the non-conduction fault optimization atlas are fused according to the fault conflict feature distribution, and the surge arrester fault first atlas is obtained.

[0080] Further, the abnormal state monitoring system of zinc oxide surge arresters in power distribution lines is also used for: according to the plurality of surge arrester neighborhoods, the a-th surge arrester neighborhood corresponding to the a-th surge arrester is extracted, and real-time action parameter collection is performed according to the a-th surge arrester neighborhood, and a-th neighborhood action data is obtained; abnormality identification is performed according to the a-th neighborhood action data, and a-th neighborhood action abnormality feature is obtained; a neighborhood interference trace architecture is activated, and the neighborhood interference trace architecture includes neighborhood interference feature-interference induced intermediate feature and interference induced basic feature; based on the a-th neighborhood action abnormality feature, the a-th surge arrester is subjected to action interference trace according to the neighborhood interference trace architecture, and a a-th neighborhood action interference path is determined; the a-th conduction accompanied fault optimization result is obtained by associating and compensating the a-th conduction accompanied fault prediction result according to the a-th neighborhood action interference path, and the a-th conduction accompanied fault optimization result is added to the conduction fault optimization atlas.

[0081] Further, the abnormal state monitoring system of zinc oxide arresters in power distribution lines is also used for: monitoring the power distribution line in real time to obtain a line monitoring data set; predicting faults of the power distribution line according to the line monitoring data set to obtain a line fault distribution; coupling fault deduction of the arrester fault first graph according to the line fault distribution based on the arrester topology distribution model to obtain a first coupled fault distribution; environment coupling fault deduction of the arrester fault first graph according to the arrester topology distribution model to obtain a second coupled fault distribution; and optimizing the arrester fault first graph according to the first coupled fault distribution and the second coupled fault distribution to generate the arrester fault second graph.

[0082] Further, the abnormal state monitoring system of zinc oxide arresters in power distribution lines is also used for: environment prediction according to the group of zinc oxide arresters to obtain an arrester environment sequence; environment anomaly attention reinforcement according to the arrester environment sequence to obtain an arrester environment vector; trend deduction of the arrester fault first graph according to the arrester environment vector based on the arrester topology distribution model to obtain a fault trend deduction result; and environment factor tracing of the fault trend deduction result according to the arrester environment vector to obtain the second coupled fault distribution.

[0083] Further, the abnormal state monitoring system of zinc oxide arresters in power distribution lines is also used for: real-time monitoring of the group of zinc oxide arresters to obtain an arrester data set; data cleaning and blockchain packaging of the arrester data set to generate the arrester monitoring data block.

[0084] Further, the abnormal state monitoring system of zinc oxide arresters in power distribution lines is also used for: generating an arrester early warning instruction according to the arrester fault second graph.

[0085] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The abnormal state monitoring method and specific examples of zinc oxide arresters in power distribution lines in the foregoing first embodiment are also applicable to the abnormal state monitoring system of zinc oxide arresters in power distribution lines in the present embodiment. Through the foregoing detailed description of the abnormal state monitoring method of zinc oxide arresters in power distribution lines, those skilled in the art can clearly understand the abnormal state monitoring system of zinc oxide arresters in power distribution lines in the present embodiment. Therefore, in order to make the specification brief, the abnormal state monitoring system of zinc oxide arresters in power distribution lines in the present embodiment is not described in detail here.

[0086] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Numerous modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without the use of the inventive faculty. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0087] It will be readily apparent to one skilled in the art that varying substitutions and modifications can be made to the application disclosed herein without departing from the scope and spirit of the application. Accordingly, it is intended that all such alterations and modifications be considered as within the scope of the application.

Claims

1. A method for monitoring abnormal states of a zinc oxide surge arrester in a power distribution line, characterized by, The method includes: Real-time monitoring of zinc oxide surge arrester groups in power distribution lines to obtain surge arrester monitoring data blocks; Based on the surge arrester monitoring data block, multi-dimensional conduction-related fault prediction is performed on the zinc oxide surge arrester group under balanced countermeasure optimization to obtain a conduction-related fault map. Based on the surge arrester monitoring data block, multidimensional non-conducting associated fault prediction is performed on the zinc oxide surge arrester group under balanced adversarial learning to obtain a non-conducting associated fault map. Based on the arrester deployment data corresponding to the zinc oxide arrester group, an arrester topology distribution model is constructed. Based on the arrester topology distribution model, neighborhood action interference compensation is performed on the conduction-accompanying fault spectrum and the non-conducting-accompanying fault spectrum to construct the first fault spectrum of the arrester. Based on the surge arrester topology distribution model, multi-level fault coupling optimization is performed on the first fault map of the surge arrester to obtain the second fault map of the surge arrester. Specifically, based on the surge arrester monitoring data block, a multi-dimensional conduction-related fault prediction method is used to optimize the zinc oxide surge arrester group under balanced countermeasures, resulting in a conduction-related fault map, including: Based on the surge arrester monitoring data block, extract the surge arrester monitoring patch corresponding to the surge arrester a, and interconnect the same type of zinc oxide surge arresters according to the surge arrester a to obtain the surge arrester cluster a, where a is a positive integer; Based on the a-th surge arrester cluster, a history retrieval of conduction-accompanying faults is performed to obtain the a-th set of conduction-accompanying faults. Based on the a-th conduction-accompanied fault history set, Q learners are trained under supervision to obtain Q conduction-accompanied fault prediction models, where Q is a positive integer greater than 1. Based on the predetermined equilibrium evaluation value, the Q conduction-accompanying fault prediction models are subjected to equilibrium adversarial optimization and aggregation to obtain the a-th conduction-accompanying fault prediction node. The monitoring module of the a-th surge arrester is input into the prediction node of the a-th conduction-accompanying fault to obtain the prediction result of the a-th conduction-accompanying fault, and the prediction result of the a-th conduction-accompanying fault is added to the conduction-accompanying fault map. Specifically, based on the surge arrester topology distribution model, neighborhood action interference compensation is performed on the conduction-related fault spectrum and the non-conducting fault spectrum to construct the first surge arrester fault spectrum, including: Based on the surge arrester topology distribution model, the neighborhood of each zinc oxide surge arrester in the zinc oxide surge arrester group is identified to obtain multiple surge arrester neighborhoods. Based on the neighborhood of the multiple surge arresters, the conduction-accompanying fault spectrum is subjected to action interference compensation to obtain the conduction fault optimization spectrum; Based on the neighborhood of the multiple surge arresters, the non-conducting fault spectrum is subjected to action interference compensation to obtain an optimized non-conducting fault spectrum. Conflict detection is performed based on the optimized continuity fault map and the optimized non-continuity fault map to obtain the fault conflict feature distribution; The optimized spectrum of conducting faults and the optimized spectrum of non-conducting faults are fused according to the fault conflict feature distribution to obtain the first spectrum of the arrester faults; Specifically, based on the surge arrester topology distribution model, multi-level fault coupling optimization is performed on the first fault map of the surge arrester to obtain the second fault map of the surge arrester, including: Real-time monitoring is performed on the power distribution line to obtain a line monitoring data set; According to the line monitoring data set, the power distribution line is predicted for failure to obtain a line failure distribution; Based on the arrester topology distribution model, the arrester fault first graph is coupled with the failure deduction according to the line failure distribution to obtain a first coupled failure distribution; According to the arrester topology distribution model, the arrester fault first graph is coupled with the failure deduction according to the line failure distribution to obtain a first coupled failure distribution; According to the first coupled failure distribution and the second coupled failure distribution, the arrester fault first graph is optimized to generate the arrester fault second graph.

2. The method of claim 1, wherein the abnormal state of the zinc oxide surge arrester in the power distribution line is monitored by the steps of: According to the predetermined balance evaluation value, the Q conduction companion fault prediction models are balanced and optimized to obtain the a conduction companion fault prediction node, including: ​ According to a plurality of conduction companion fault indicators, the a conduction companion fault history set is classified to obtain a plurality of conduction companion fault partitions; Sample balance evaluation is performed on the plurality of conduction companion fault partitions to obtain a sample balance evaluation value; According to the predetermined balance evaluation value, the sample balance evaluation value is difference identified to obtain a balance evaluation difference feature; According to the plurality of conduction companion fault partitions, the balance evaluation difference feature is traced to determine a balance difference trace factor; According to the balance difference trace factor, the plurality of conduction companion fault partitions are injected with counter samples to obtain the a conduction companion fault optimization set; According to the a conduction companion fault optimization set, the Q conduction companion fault prediction models are tested to obtain a Q conduction companion fault loss set; According to the Q conduction companion fault loss set, the Q conduction companion fault prediction models are optimized and aggregated to generate the a conduction companion fault prediction node.

3. The method for monitoring the abnormal state of zinc oxide surge arresters in power distribution lines as described in claim 1, characterized in that, According to the plurality of arrester neighborhoods, the conduction companion fault graph is compensated for action intervention to obtain a conduction fault optimization graph, including: According to the plurality of arrester neighborhoods, the a arrester is extracted from the a arrester neighborhood, and real-time action parameter collection is performed according to the a arrester neighborhood to obtain a a neighborhood action data; According to the a neighborhood action data, an abnormality is identified to obtain a a neighborhood action abnormality feature; The neighborhood intervention trace architecture is activated, and the neighborhood intervention trace architecture includes neighborhood intervention features, intervention-induced intermediate features, and intervention-induced basic features; Based on the a neighborhood action abnormality feature, the a arrester is traced for action intervention according to the neighborhood intervention trace architecture to determine a a neighborhood action intervention path; According to the a neighborhood action intervention path, a a conduction companion fault prediction result is associated and compensated to obtain a a conduction companion fault optimization result, and the a conduction companion fault optimization result is added to the conduction fault optimization graph.

4. The method of claim 1, wherein the abnormal state of the zinc oxide surge arrester in the power distribution line is monitored by using the voltage and the current of the power distribution line. According to the arrester topology distribution model, the arrester fault first graph is coupled with the failure deduction according to the line failure distribution to obtain a first coupled failure distribution, including: According to the zinc oxide arrester group, an environment prediction is performed to obtain an arrester environment sequence; According to the environment sequence of each arrester, environment anomaly attention reinforcement is performed to obtain an environment vector of each arrester; Based on the arrester topology distribution model, trend deduction is performed on the arrester fault first graph according to the environment vector of each arrester to obtain a fault trend deduction result of each arrester; According to the environment vector of each arrester, environment factor tracing is performed on the fault trend deduction result of each arrester to obtain the second coupled fault distribution.

5. The method of claim 1, wherein the abnormal state of the zinc oxide surge arrester in the power distribution line is monitored by using the voltage and the current of the power distribution line. Real-time monitoring of a zinc oxide arrester group of a power distribution line is performed to obtain an arrester monitoring data block, including: Real-time monitoring of the zinc oxide arrester group is performed to obtain an arrester data set; Data cleaning and blockchain packaging are performed on the arrester data set to generate the arrester monitoring data block.

6. The method of claim 1, wherein the abnormal state of the zinc oxide surge arrester in the power distribution line is monitored by using a method of detecting a change in a current flowing through the zinc oxide surge arrester. According to the arrester fault second graph, an arrester early warning instruction is generated.

7. A system for monitoring abnormal conditions of zinc oxide surge arresters in power distribution lines, characterized by Steps of an abnormal state monitoring method of a zinc oxide arrester in a power distribution line according to any one of claims 1 to 6, including: An arrester monitoring data block obtaining module is configured to perform real-time monitoring of a zinc oxide arrester group of a power distribution line to obtain an arrester monitoring data block; A conduction accompanying fault graph obtaining module is configured to perform multi-dimensional conduction accompanying fault prediction of the zinc oxide arrester group under balanced confrontation optimization according to the arrester monitoring data block to obtain a conduction accompanying fault graph; A non-conduction accompanying fault graph obtaining module is configured to perform multi-dimensional non-conduction accompanying fault prediction of the zinc oxide arrester group under balanced confrontation learning according to the arrester monitoring data block to obtain a non-conduction accompanying fault graph; An arrester topology distribution model constructing module is configured to construct an arrester topology distribution model according to arrester layout data corresponding to the zinc oxide arrester group; An arrester fault first graph constructing module is configured to perform neighborhood action interference compensation on the conduction accompanying fault graph and the non-conduction accompanying fault graph according to the arrester topology distribution model to construct an arrester fault first graph; An arrester fault second graph obtaining module is configured to perform multi-level fault coupling optimization on the arrester fault first graph according to the arrester topology distribution model to obtain an arrester fault second graph.

Images