An agent-based partial discharge monitoring operation and maintenance method, system and device

Abstract

The application relates to the technical field of partial discharge monitoring, in particular to a partial discharge monitoring operation and maintenance method, system and device based on an intelligent agent, which comprises the following steps: periodically collecting target data reported by at least one edge side device, the target data being determined based on at least one power grid terminal covered by the edge side device, and the target data comprising monitoring data and pre-diagnosis data; when the target data is obtained in each period, performing data analysis on the target data by using a data analysis model to obtain real-time analysis results; for each edge side device, gathering the real-time analysis results and the target data in an analysis interval to obtain a data group; and analyzing the data group by using the data analysis model to obtain a partial discharge diagnosis result, wherein the partial discharge diagnosis result comprises an abnormal probability of a partial discharge anomaly existing in the power grid terminal and a partial discharge type. The method can realize intelligent evaluation, diagnosis and risk prediction of a partial discharge state of a power grid terminal, and the diagnosis is timely and highly accurate.

Description

Technical Field

[0001] This application relates to the field of partial discharge monitoring technology, and in particular to a partial discharge monitoring operation and maintenance method, system and device based on intelligent agents. Background Technology

[0002] The operation and maintenance of power systems are crucial to ensuring the stable and efficient operation of power equipment and related facilities over long periods. This mainly includes regular operational monitoring, maintenance, and fault repair to ensure the safe, stable, and reliable operation of the power system.

[0003] The typical operation and maintenance method is manual periodic testing, which mainly relies on maintenance personnel to regularly visit various sites to inspect and record the equipment one by one.

[0004] However, manual inspection is not highly automated and is inefficient. In particular, the detection cycle for partial discharge diagnosis of equipment such as switchgear in the power distribution room is too long and the accuracy is low, which can easily lead to the failure to detect potential partial discharge faults in a timely manner. Summary of the Invention

[0005] This application provides a partial discharge monitoring and maintenance method, system, and device based on intelligent agents to solve the problems of low efficiency and accuracy of traditional monitoring methods.

[0006] In a first aspect, embodiments of this application provide a partial discharge monitoring and maintenance method based on an intelligent agent, applied to a cloud intelligent agent. The method includes: periodically collecting target data reported by at least one edge-side device, the target data being determined based on at least one power grid terminal covered by the edge-side device, the target data including monitoring data and pre-diagnostic data, the pre-diagnostic data being generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data; when target data is acquired in each cycle, data analysis is performed on the target data using a data analysis model to obtain real-time analysis results; for each edge-side device, the real-time analysis results and target data within an analysis interval are aggregated to obtain a data set; wherein, an analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, M and N being preset values; the data set is analyzed using a data analysis model to obtain partial discharge diagnosis results, the partial discharge diagnosis results including the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0007] In one possible implementation, the monitoring data includes partial discharge monitoring data; the steps of using a data analysis model to analyze the target data and obtain real-time analysis results include: plotting the partial discharge monitoring data to obtain an analysis spectrum; the analysis spectrum includes a partial discharge pulse signal spectrum; using a spectrum classification and recognition model to perform waveform analysis on the partial discharge pulse signal spectrum to determine the waveform characteristics of the analysis spectrum; wherein, the waveform characteristics include one or more of waveform type, peak information, and trough information, and the waveform type includes single peak, double peak, multi-peak, and stable peak; the real-time analysis results include waveform characteristics.

[0008] In one possible implementation, the analysis spectrum also includes a partial discharge phase analysis spectrum. After plotting the partial discharge monitoring data to obtain the analysis spectrum, the method further includes: using a mechanism model to perform feature calculations on the partial discharge pulse signal spectrum to obtain analysis features, including amplitude dispersion features, pulse uniformity features, polarity features, phase clustering features, 50Hz component significance features, 100Hz component significance features, and 50 / 100Hz component numerical relationship features; using a mechanism model to perform feature calculations on the partial discharge phase analysis spectrum to obtain flight pattern features; the real-time analysis results also include the analysis features and flight pattern features.

[0009] In one possible implementation, the pre-diagnostic data includes partial discharge alarm information; the steps of analyzing the data set using a data analysis model to obtain the partial discharge diagnosis results include: for each edge-side device, determining whether there is a discharge interference signal in the power grid terminal it covers based on the data set; if there is no discharge interference signal, determining the partial discharge type as no anomaly, and determining the anomaly probability as 0; if there is a discharge interference signal, determining whether the number of partial discharge alarm information in the data set is greater than a preset threshold; if the number of partial discharge alarm information in the data set is not greater than the preset threshold, determining the partial discharge type as no anomaly, and determining the anomaly probability as 0.

[0010] In one possible implementation, the pre-diagnostic data includes partial discharge alarm information; the partial discharge type includes target partial discharge type and particulate discharge type; the step of using a data analysis model to analyze the data set and obtain the partial discharge diagnosis result further includes: determining whether the number of partial discharge alarm messages in the data set is greater than a preset threshold; if the number of partial discharge alarm messages in the data set is greater than the preset threshold, retrieving a rule table from the database; the rule table includes signal characteristic information corresponding to different target partial discharge types; the target partial discharge types include, but are not limited to, air gap discharge type, surface discharge type, suspended discharge type, and corona discharge type; and determining the characteristics of each data set according to the rule table. The degree of matching between the waveform features and analytical features of the corresponding analysis spectrum and the signal feature information of each target partial discharge type in the rule table is determined, and the feature probability of each waveform feature and each analytical feature is determined based on the judgment result. Specifically, for each target partial discharge type, if the waveform features or analytical features of the analysis spectrum corresponding to the data group match the corresponding signal feature information, the feature probability corresponding to that waveform feature or analytical feature is 1; if it does not match the signal feature information, the feature probability corresponding to that waveform feature or analytical feature is 0. The type probability P corresponding to each target partial discharge type for each data group is calculated based on the first formula, which is: Where Xi represents the feature probability corresponding to the waveform feature or analysis feature in the data set, q equals the total number of waveform features and analysis features in the data set, 0 < i ≤ q; if the waveform features and analysis features of the analysis spectrum corresponding to the data set do not match the signal feature information corresponding to each target partial discharge type, the similarity of the flight map features is used to determine whether the flight map features present significant features; where the flight map features present significant features and non-significant features correspond to different flight map feature similarities; if the flight map features present significant features, the similarity of the flight map features corresponding to the significant features is determined as the type probability corresponding to the particle discharge type; based on the type probability of each data set corresponding to each partial discharge type, the partial discharge diagnosis result is obtained.

[0011] In one possible implementation, the step of obtaining the partial discharge diagnosis result based on the type probability of each data group corresponding to each partial discharge type includes: for each data group, determining the maximum probability value among the type probabilities of the data group corresponding to each partial discharge type, and counting the number of type probabilities equal to the maximum probability value; if the number of type probabilities equal to the maximum probability value is one, determining that the partial discharge type corresponding to the data group is the target partial discharge type or particulate discharge type corresponding to the maximum probability value, and determining that the abnormal probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge-side device corresponding to the data group; if the number of type probabilities equal to the maximum probability value is multiple, determining that the partial discharge type corresponding to the data group is a random one of the target partial discharge type or particulate discharge type corresponding to the maximum probability value, and determining that the abnormal probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge-side device corresponding to the data group.

[0012] In one possible implementation, the monitoring data also includes environmental monitoring data; the method further includes: aggregating the partial discharge type and anomaly probability corresponding to each edge-side device to form a monitoring list; and / or, for each edge-side device, generating disposal suggestions based on the partial discharge type, anomaly probability, partial discharge monitoring data, environmental monitoring data, and / or the ledger data of the power grid terminal, wherein the ledger data records the basic information, operating status, and / or maintenance history of the power grid terminal; and / or, obtaining the judgment results for the partial discharge diagnosis results to form positive samples and negative samples; positive samples correspond to judgment results that match the partial discharge diagnosis results, and negative samples correspond to judgment results that do not match the partial discharge diagnosis results; and using the positive and negative samples to train a data analysis model.

[0013] In one possible implementation, the method further includes: responding to a text question submitted by a user through a web page, invoking a large text model to parse the question and generate an answer; wherein the large text model is trained based on a preset training document, which includes partial discharge monitoring data, environmental monitoring data, ledger data of power grid terminals, and / or partial discharge knowledge; and displaying the answer on the web page.

[0014] Secondly, this application also provides a partial discharge monitoring and maintenance system based on an intelligent agent. The system includes: edge-side devices and a cloud intelligent agent; the edge-side devices are configured to: generate target data and report it to the cloud intelligent agent; the target data is determined based on at least one power grid terminal covered by the edge-side devices, and the target data includes monitoring data and pre-diagnostic data, the pre-diagnostic data being generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data; the cloud intelligent agent is configured to: periodically collect the target data reported by at least one edge-side device; when the target data is acquired in each cycle, the target data is analyzed using a data analysis model to obtain real-time analysis results; for each edge-side device, the real-time analysis results and target data within an analysis interval are aggregated to obtain a data set; wherein, an analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, where M and N are preset values; the data set is analyzed using a data analysis model to obtain partial discharge diagnosis results, the partial discharge diagnosis results including the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0015] Thirdly, this application also provides a partial discharge monitoring and maintenance device based on an intelligent agent, applied to a cloud intelligent agent. The device includes: a data acquisition module configured to periodically acquire target data reported by at least one edge-side device, the target data being determined based on at least one power grid terminal covered by the edge-side device, the target data including monitoring data and pre-diagnostic data, the pre-diagnostic data being generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data; a first analysis module configured to perform data analysis on the target data using a data analysis model when the target data is acquired in each cycle, to obtain real-time analysis results; an aggregation module configured to aggregate the real-time analysis results and target data within an analysis interval for each edge-side device, to obtain a data set; wherein, an analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, M and N being preset values; and a second analysis module configured to analyze the data set using a data analysis model to obtain partial discharge diagnosis results, the partial discharge diagnosis results including the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0016] As described above, this application provides a method, system, and apparatus for partial discharge monitoring and maintenance based on intelligent agents. The method includes: periodically collecting target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device, and includes monitoring data and pre-diagnostic data. When target data is acquired in each cycle, a data analysis model is used to analyze the target data to obtain real-time analysis results. For each edge-side device, the real-time analysis results and target data within an analysis interval are aggregated to obtain a data set. The data set is analyzed using the data analysis model to obtain partial discharge diagnosis results, which include the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge. This method can achieve intelligent assessment, diagnosis, and risk prediction of the partial discharge status of power grid terminals, with timely and high accuracy in diagnosis. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the first process of the partial discharge monitoring and maintenance method based on intelligent agents provided in the embodiments of this application;

[0018] Figure 2 This is a second flowchart illustrating the agent-based partial discharge monitoring and maintenance method provided in this application embodiment.

[0019] Figure 3 A flowchart illustrating the pre-diagnosis steps provided in an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of the intelligent question-answering process provided in the embodiments of this application;

[0021] Figure 5 This is a schematic diagram of the first structure of the agent-based partial discharge monitoring and maintenance system provided in the embodiments of this application;

[0022] Figure 6 This is a schematic diagram of a second structure of the agent-based partial discharge monitoring and maintenance system provided in an embodiment of this application;

[0023] Figure 7 This is a schematic diagram of the structure of the partial discharge monitoring and maintenance device based on intelligent agents provided in the embodiments of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0025] Before introducing the technical solutions of the embodiments of this application, the terminology involved in the embodiments of this application will be introduced by way of example.

[0026] 1. Intelligent Agent: An entity that can autonomously perceive its environment, process information, perform tasks, and interact with the outside world.

[0027] 2. Power Distribution Station / Room: A physical building housing transformers (optional), switchgear, and other electrical equipment and incoming / outgoing cables. It is equipped with lighting, fire protection, and drainage facilities. The power distribution station / room connects to the substation above and the user-side distribution room below, primarily fulfilling two functions: first, medium-voltage to low-voltage substation operation; second, organizing the power distribution network through busbars and switchgear, supporting flexible power allocation. Power distribution stations / rooms are mainly divided into two major categories and four subcategories:

[0028] Switching stations and switchgear (indoors called boundary rooms, outdoor ones called ring main units): realize distribution network organization and load distribution, generally do not act as substations, and do not directly connect to end users.

[0029] Distribution room and transformer substation: Participate in the distribution network organization to achieve ring network power supply; realize 10kV / 400V substation to provide power to the load.

[0030] 3. Switch cabinet: A type of equipment used in power systems, primarily for the distribution, control, protection, and monitoring of electrical energy. A switch cabinet includes a series of electrical devices such as circuit breakers, disconnect switches, load switches, grounding switches, transformers, protective relays, and control devices.

[0031] 4. Partial Discharge: Partial discharge is a transient discharge process that occurs in a localized area inside or on the surface of the insulation structure of electrical equipment. It is a sign of insulation degradation. Partial discharge does not immediately cause insulation failure, but it will accelerate the degradation. If left untreated, it will eventually lead to insulation breakdown and cause an accident.

[0032] The operation and maintenance of power systems are crucial to ensuring the stable and efficient operation of power equipment and related facilities over long periods. This mainly includes regular operational monitoring, maintenance, and fault repair to ensure the safe, stable, and reliable operation of the power system.

[0033] The typical operation and maintenance method is manual periodic testing, which mainly relies on maintenance personnel to regularly visit various sites to inspect and record the equipment one by one.

[0034] However, manual inspection is not highly automated and is inefficient. In particular, the detection cycle for partial discharge diagnosis of equipment such as switchgear in the power distribution room is too long and the accuracy is low, which can easily lead to the failure to detect potential partial discharge faults in a timely manner.

[0035] Partial discharge is usually caused by internal defects in the insulation structure or changes in external factors that lead to an excessively large local electric field, triggering the discharge.

[0036] Internal factors: gaps, cracks, bubbles, dirt, and uneven density inside or on the surface of the insulation.

[0037] External factors: mechanical vibration / frequent switching leading to poor conductor contact, excessively high temperature leading to reduced insulation performance, excessively high humidity leading to condensation on the insulation surface, overvoltage and overcurrent leading to excessive local electric field, and unreasonable structure.

[0038] This application provides a partial discharge monitoring and maintenance method, system, and device. This method can realize online monitoring of power grid terminals, perform early monitoring and warning of problems, and reduce the workload of front-line live-line testing. At the same time, it can detect equipment abnormalities in the first instance, avoid the randomness and lag of periodic live-line testing, and reduce the risk of equipment safety operation.

[0039] Figure 1 This is a schematic diagram of the first process of the partial discharge monitoring and maintenance method based on intelligent agents provided in the embodiments of this application.

[0040] Figure 2 This is a second flowchart illustrating the partial discharge monitoring and maintenance method based on intelligent agents provided in the embodiments of this application.

[0041] like Figure 1 and Figure 2 As shown, the partial discharge monitoring and maintenance method based on intelligent agents provided in this application embodiment can be applied to cloud intelligent agents, and may specifically include the following steps S100-S400.

[0042] S100: Periodically collect target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device. The target data includes monitoring data and pre-diagnostic data. The pre-diagnostic data is generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data.

[0043] The power grid terminal can be a switchgear in a substation. Monitoring data can be obtained by periodically collecting data from the switchgear and other power grid terminals through sensing terminals. Monitoring data can include partial discharge monitoring data and environmental monitoring data. Furthermore, the edge-side equipment can be a station-end node; one station-end node can manage one substation and its associated sensing terminals.

[0044] In some implementations, multiple switchgear cabinets can be arranged in the substation room. The actual number of switchgear cabinets and the number of substation rooms depend on the actual power system layout, and this application embodiment does not specifically limit this. The sensing terminal may include a partial discharge sensor and a temperature and humidity sensor. Each substation room is equipped with a temperature and humidity sensor to collect environmental monitoring data of the substation room, such as temperature and humidity. Furthermore, each switchgear cabinet in each substation room is equipped with a partial discharge sensor to sample partial discharge signals or discharge interference signals.

[0045] In some implementations, the sensing terminal can specifically collect data from the switch cabinets in the substation room, or from the incoming line cabinets, feeder cabinets, bus tie cabinets, bus tie isolation cabinets, voltage transformer (PT) cabinets, and transformers within the substation (referred to as substation transformers).

[0046] It should also be noted that pre-diagnosis is edge-level computation and can be implemented by edge-side devices. The steps of pre-diagnosis may include locating and eliminating external interference, generating station-end partial discharge and interference alarms. The specific steps will be detailed below and will not be repeated here. This can reduce the computing pressure on the cloud.

[0047] Furthermore, in the embodiments of this application, the frequency of periodic collection can be preset, such as 2h / time or 4h / time, and the embodiments of this application do not specifically limit it.

[0048] In some implementations, edge devices and cloud agents can communicate via wireless LAN authentication and privacy infrastructure (WAPI), carrier 5G, or wired Ethernet networks.

[0049] S200: When the target data is acquired in each cycle, the data analysis model is used to analyze the target data and obtain real-time analysis results.

[0050] This application embodiment can perform real-time partial discharge analysis based on pre-diagnostic data and monitoring data. The data analysis model, also known as the partial discharge diagnostic model, can include an AI graph classification model (also known as a graph classification and recognition model), a mechanism feature analysis model (also known as a mechanism model), etc. Specifically, the AI ​​graph classification model can be a CNN classification model (also known as a partial discharge graph CNN classification model). It is understood that step S200 is a single sampling analysis calculation step, which can include AI graph classification and recognition and mechanism model feature calculation.

[0051] In this way, the target data reported by each edge device can be analyzed in real time to perform real-time partial discharge diagnosis on the power grid terminals covered by the edge device and detect partial discharge faults in a timely manner.

[0052] S300: For each edge device, it aggregates real-time analysis results and target data within an analysis interval to obtain a data set.

[0053] Here, an analysis interval refers to the interval corresponding to the Mth acquisition period to the (M+Nth)th acquisition period, where M and N are preset values. For example, an analysis interval can refer to the interval corresponding to the 1st acquisition period to the 14th acquisition period, in which case M equals 1 and N equals 13. In this way, the acquisition time over a period of time can be summarized for further partial discharge diagnosis. It can be understood that step S300 may include the aggregation of AI map classification results and the aggregation of mechanism model partial discharge type analysis results.

[0054] S400: Analyzes the data set using a data analysis model to obtain partial discharge diagnosis results, including the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0055] The partial discharge diagnosis results can include the partial discharge type (PD Type) and probability of anomalies in the power grid terminals. For example, the partial discharge diagnosis results can include the partial discharge type and probability of anomalies in the switchgear in the substation.

[0056] Furthermore, different defects lead to different partial discharge types, which can include targeted partial discharge types and particulate discharge types. Target partial discharge types can include air gap discharge types, surface discharge types, suspended discharge types, and corona discharge types. The anomaly probability is obtained based on data analysis model analysis; the specific analysis steps will be detailed below and will not be repeated here.

[0057] In some implementations, step S400 can also perform partial discharge diagnosis based on historical data collected from the power grid terminal. Historical data can refer to data collected from the same power grid terminal on the previous day or several days prior.

[0058] In this way, through the collaboration of cloud, edge, and terminal, real-time monitoring of partial discharge anomalies in power grid terminals (such as switchgear in substation rooms) can be achieved. Anomaly identification is accurate and timely, while saving manpower, improving equipment operation stability, and ensuring the safe operation of power grid terminals.

[0059] As described above, this application provides a partial discharge diagnosis and monitoring method based on intelligent agents. The method includes: periodically collecting target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device. The target data includes monitoring data and pre-diagnostic data, which is generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data. When target data is acquired in each cycle, a data analysis model is used to analyze the target data to obtain real-time analysis results. For each edge-side device, the real-time analysis results and target data within an analysis interval are aggregated to obtain a data set. An analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, where M and N are preset values. The data set is analyzed using a data analysis model to obtain partial discharge diagnosis results, which include the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge. This method can achieve intelligent assessment, diagnosis, and risk prediction of the partial discharge status of power grid terminals, realizing a shift from "diagnosing problems based on personal experience" to "intelligent diagnosis based on model reasoning," resulting in fewer false alarms, faster diagnosis, and higher reliability of partial discharge results.

[0060] It should also be noted that in traditional partial discharge detection and diagnosis methods, users mainly rely on a single on-site live-line test to diagnose the presence of partial discharge. The results of a single test are easily affected by occasional factors, leading to misjudgment. Secondly, the diagnosis of partial discharge relies on personal experience, which may also be affected by individual skills and personnel changes, leading to misjudgment. To avoid misjudgment, users must continuously perform multiple on-site retests, which is inefficient.

[0061] After the agent-based method and system provided in this application are put into use, firstly, they can achieve comprehensive diagnosis of multiple partial discharge monitoring results based on historical online monitoring data, avoiding the influence of incidental factors and improving the reliability of diagnosis; secondly, by combining mechanistic models with AI models, they can transform purely manual diagnosis into intelligent system diagnosis, avoiding the limitations of individual abilities and experience and improving the accuracy of diagnosis; furthermore, the correlation analysis and diagnosis of general partial discharge data with temperature, humidity, load, and record data improves the accuracy of response measures. Thirdly, relying on continuous online monitoring and diagnosis, the workload of on-site retesting is reduced, and the efficiency of partial discharge diagnosis is improved.

[0062] The following section details the steps for pre-diagnosing edge-side devices.

[0063] Figure 3 This is a flowchart illustrating the pre-diagnosis steps provided in an embodiment of this application.

[0064] like Figures 1 to 3 As shown, the partial discharge diagnostic monitoring method based on intelligent agents provided in this application embodiment may further include the following steps S501-S504.

[0065] S501: Receives the collected data reported by the sensing terminal, performs time alignment based on the timestamp carried by the collected data, and obtains all collected data within the same time window. The same time window corresponds to one collection cycle.

[0066] Understandably, when sampling signals and generating collected data, the sensing terminal can add timestamps to the collected data to indicate the specific sampling time. Since the sensing terminal samples periodically, all collected data for a substation within a single collection cycle can be determined based on the timestamps. The key to time alignment is to unify these different collected data into a common time standard. This ensures that all data are within the same time window, achieving time synchronization and facilitating data analysis.

[0067] S502: Identify the collected data within the same time window, determine whether it includes partial discharge monitoring data of all power grid terminals covered by the edge-side equipment, and / or determine whether there are any abnormal values.

[0068] Understandably, this step involves checking the data within the current time window, first determining whether it includes partial discharge monitoring data for all power grid terminals that need to be monitored. Partial discharge monitoring data is one of the key data used to assess whether potential faults exist in power grid terminals. If the data lacks monitoring results from certain devices, this embodiment can mark these missing data portions. Furthermore, this embodiment can also perform outlier detection on the partial discharge monitoring data. For example, some values ​​in the partial discharge monitoring data may deviate due to sensor malfunctions, communication problems, or other interference factors. In this case, this embodiment can identify these outliers for subsequent correction.

[0069] S503: If partial discharge monitoring data for all grid terminals covered by edge-side equipment is not included, supplement the missing partial discharge monitoring data, and / or, if outliers are included, correct the outliers.

[0070] In this embodiment, if partial discharge monitoring data for certain power grid terminals (e.g., switchgear) is missing, numerical supplementation calculations can be performed based on data from other power grid terminals or historical partial discharge monitoring data of that terminal. Furthermore, this can be achieved through various methods, such as model-based prediction or extrapolation from monitoring data of adjacent devices. The purpose of supplementation is to ensure that partial discharge monitoring data from all power grid terminals are fully reflected within the current time window.

[0071] For data marked as anomalous, embodiments of this application can perform anomaly correction. For example, erroneous data can be replaced with reasonable values ​​through data filtering, calibration algorithms, and other methods. In this way, the corrected data will more accurately reflect the true state of the device.

[0072] S504: Uses ultrasonic algorithms to perform pre-diagnosis on supplemented and / or corrected partial discharge monitoring data within the same time window. When a target signal is identified, pre-diagnosis data is generated.

[0073] The target signal includes partial discharge (PD) signals and / or external discharge interference signals. The pre-diagnostic data includes PD alarm information and / or interference alarm information. The PD alarm information includes at least the coordinates corresponding to the occurrence of the PD signal, and the interference alarm information includes at least the coordinates corresponding to the occurrence of the PD interference signal. In this embodiment, "coordinates" can refer to the precise location information of the PD signal or PD interference signal in three-dimensional space. Specifically, it can be the specific physical location of the power grid terminal (such as a switchgear) where the PD signal or PD interference signal occurs, or the specific physical location of the PD sensor that captures the PD signal or PD interference signal. "Coordinates" can include coordinate values ​​in the X, Y, and Z dimensions.

[0074] In some implementations, "coordinates" can refer to the precise location information of partial discharge signals or partial discharge interference signals in the time dimension, specifically the timestamp when the partial discharge signal or partial discharge interference signal is captured.

[0075] In this embodiment, the partial discharge monitoring data is formed from the partial discharge monitoring signal. The partial discharge monitoring signal can be a pulse. For example, the pulse is formed by sampling the superimposed signals of partial discharge signals and / or external interference signals in the substation room during the acquisition period. Therefore, when partial discharge signals and / or external interference signals are detected in the partial discharge monitoring signal, it can be determined that there is an anomaly in the switchgear in the substation room, thus achieving the purpose of external interference marking, alarm reporting, and data reporting.

[0076] In some implementations, when the partial discharge monitoring signal exhibits partial discharge signal characteristics as a pulse, or exhibits interference signal characteristics, the target signal can be identified in the partial discharge monitoring signal.

[0077] Furthermore, if a target signal is detected, embodiments of this application can generate partial discharge alarm information based on the characteristics of the signal. The partial discharge alarm information may include the source location of the signal (such as the equipment coordinates of a specific power grid terminal) and the time of occurrence, thereby helping maintenance personnel to locate faulty equipment.

[0078] Furthermore, the ultrasonic algorithm can also distinguish partial discharge signals from other possible interference signals. Interference signals outside the cabinet may be caused by external environmental factors (such as electrical noise, wind noise, etc.). The ultrasonic algorithm can identify the presence of partial discharge signals or interference signals by analyzing the frequency and amplitude differences of the signals. Partial discharge signals typically have specific pulse characteristics, while interference signals are usually continuous noise. The embodiments of this application can identify these interference signals and generate interference alarm information, recording the coordinates of the partial discharge interference signals for further analysis and elimination of false alarms.

[0079] The following section details the steps involved in cloud-based intelligent agent data analysis.

[0080] In this embodiment, the cloud agent can invoke a data analysis model to perform analysis and calculations based on nine mechanistic characteristics of partial discharge monitoring data to obtain real-time analysis results. The nine mechanistic characteristics are as follows:

[0081] ① Amplitude dispersion: Analyze the amplitude range of the signal on each phase to determine its dispersion, and combine it with different partial discharge types to determine its characteristic conformity. For example, the amplitude dispersion of suspended discharge is small.

[0082] ② Pulse equalization: Analyze and count the pulse frequencies (meeting certain amplitude intensities) within each power frequency cycle of the sampled data to obtain a sequence of 50 pulse frequencies from one sampling; analyze the equalization based on the pulse count of this sequence.

[0083] ③ Phase spectrum waveform feature analysis: Analyze the peak and valley morphology of the phase spectrum, analyze the single-peak, double-peak, and multi-peak characteristics, and judge the characteristic conformity in combination with different partial discharge types. For example, floating discharge shows double-peak characteristics.

[0084] ④ Polarity characteristics: Analyze and calculate whether the peak and valley shapes of the phase spectrum conform to the polarity effect.

[0085] ⑤ Phase clustering characteristics: Analyze and calculate the pulse clustering in the phase spectrum to determine whether the signal is concentrated on a small number of phases or is highly dispersed, such as high clustering of suspended discharge.

[0086] ⑥ Significance of the 50Hz component: The significance of the 50Hz component is evaluated by comparing the peak amplitude of the signal with the 50Hz frequency component of the ultrasonic signal.

[0087] ⑦ 100Hz component significance: The significance of the 100Hz component of the ultrasonic signal is evaluated by comparing it with the peak amplitude of the signal.

[0088] ⑧ 50 / 100Hz component numerical relationship (or 50-100Hz numerical relationship characteristics): Analyze the data magnitude relationship between the two, and judge the characteristic conformity in combination with different discharge types. For example, the 50Hz component of the floating discharge is significantly smaller than the 100Hz component.

[0089] ⑨ Flight pattern features: used to characterize particle discharge.

[0090] Furthermore, the real-time analysis results output by the cloud intelligent agent can include waveform features, analytical features, and / or flight map features. Waveform features can be single-peak, double-peak, multi-peak, or stationary peaks, and can be determined based on phase spectrum waveform feature analysis. Analytical features can include amplitude dispersion features, pulse equalization features, polarity features, phase clustering features, 50Hz component significance features, 100Hz component significance features, and 50 / 100Hz component numerical relationship features, totaling seven types.

[0091] like Figures 1 to 3 As shown, step S200 may include the following steps S201-S202.

[0092] S201: Plot the partial discharge monitoring data to obtain the analysis spectrum.

[0093] The analysis graphs can include partial discharge pulse signal (PRPS) graphs and partial discharge phase analysis (PRPD) graphs. The PRPS graph presents the amplitude, phase, and time period of the partial discharge monitoring signal in a three-dimensional format. The PRPD graph typically combines 50 power frequency cycles of sampled data, categorized by phase from 0 to 360°, counts the same amplitude for the same phase, and presents the data as a scatter plot (different counts correspond to different colors). In practical applications, the graph can be plotted based on the sampled data (sampled amplitude) within a complete electrical cycle (360°) of the PRPS data. Typically, there are 3600 sampled amplitudes within a complete electrical cycle, i.e., 3600 sampling points. 3600 sampling points ensure at least one sampling point for every 0.1° phase, meeting the accuracy requirements for graph plotting.

[0094] S202: Use a spectrum classification and recognition model to perform waveform analysis on the spectrum of partial discharge pulse signals in order to determine the waveform characteristics of the spectrum; wherein, the waveform characteristics include one or more of waveform type, peak information, and trough information, and the waveform type includes single peak, double peak, multi-peak, and steady peak.

[0095] This step is the calculation of the waveform characteristics of the phase spectrum. The peak information and valley information may specifically include the effective peak and its left and right adjacent valleys, the start phase, the end phase, and the peak value, etc. This application embodiment does not specifically limit this.

[0096] In this embodiment, the graph classification and recognition model can be obtained by optimizing a basic model based on an AI classification model architecture (such as SVM, CNN, etc.) for partial discharge scenarios. During the training phase, PRPS data containing waveforms such as single-peak and double-peak waveforms are first preprocessed. The preprocessing steps include: data denoising, amplitude standardization, normalization, and labeling waveform types and / or peak / trough features (such as phase range, peak value, etc.). Then, the preprocessed PRPS data is used as training data to perform supervised learning training on the basic model, ultimately obtaining a model capable of accurately recognizing waveform features. Furthermore, step S201 may be followed by steps S203-S204.

[0097] S203: The partial discharge pulse signal spectrum is analyzed by using a mechanism model to obtain the analytical features.

[0098] S204: The characteristics of the partial discharge phase analysis spectrum are calculated using the mechanism model to obtain the flight pattern characteristics.

[0099] It is worth noting that the embodiments of this application can classify various features. For example, amplitude dispersion can be divided into high, medium, and low types; pulse equalization can also be divided into high, medium, and low types; phase spectrum waveform features can be divided into single-peak, double-peak, multi-peak, and stable peak; polarity features can be divided into inconspicuous, relatively significant, and significant types; phase aggregation can be divided into high, medium, and low types; the 50Hz component can be divided into high, medium, and low types; the 100Hz component can be divided into high, medium, and low types; and the 50 / 100Hz component numerical relationship features can include 50Hz component >> 100Hz component or 100Hz component >> 50Hz component, where ">>" is a much greater than sign. Flight pattern features include similarity characterization and flight pattern feature similarity. Similarity characterization includes at least the presence of significant features and the presence of insignificant features. Different flight pattern feature similarities correspond to different flight pattern feature similarities when significant features and insignificant features are presented. Significant flight pattern features can indicate the presence of particle discharge, while insignificant features can indicate the absence of particle discharge. Furthermore, step S400 may include the following steps S401-S404.

[0100] S401: For each edge-side device, determine whether there is a discharge interference signal at the power grid terminal it covers based on the data set.

[0101] In this embodiment of the application, the step of determining whether there is an interference signal based on the data group can specifically be to determine whether the pre-diagnostic data includes partial discharge alarm information for interference signals.

[0102] S402: If there is no discharge interference signal, the partial discharge type is determined to be no abnormality, and the probability of abnormality is determined to be 0.

[0103] In this way, it can be determined that there is no partial discharge anomaly in the power grid terminal covered by the edge-side equipment.

[0104] In this embodiment of the application, if it is determined that there is no partial discharge anomaly, routine monitoring can be resumed and step S100 can continue to be executed.

[0105] S403: If there is a discharge interference signal, determine whether the number of partial discharge alarm messages in the data group is greater than the preset threshold.

[0106] The preset threshold is, for example, equal to 1 or 2, but this application embodiment does not specifically limit it.

[0107] S404: If the number of partial discharge alarm messages in the data group is not greater than the preset threshold, the partial discharge type is determined to be without anomaly, and the probability of anomaly is determined to be 0.

[0108] In this way, it can be determined that there is no partial discharge anomaly in the power grid terminal covered by the edge-side equipment.

[0109] Furthermore, step S400 may also include the following steps S405-S407.

[0110] S405: If the number of partial discharge alarm messages in the data group is greater than the preset threshold, retrieve the rule table from the database; the rule table includes signal characteristic information corresponding to different target partial discharge types; the target partial discharge types include but are not limited to air gap discharge type, surface discharge type, suspension discharge type and / or corona discharge type.

[0111] For example, the rule table is shown in Table 1.

[0112] Table 1 Rules Table

[0113]

[0114] In this table, the rows represent various mechanistic characteristics, the columns represent various partial discharge types, and the cells composed of rows and columns show the signal characteristic information corresponding to various partial discharge types.

[0115] In some implementations, the rule table can be stored in a database for quick rule modifications. During each diagnosis, the database is read, and probability calculations are performed according to the latest rules.

[0116] S406: Based on the rule table, determine the degree of matching between the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to each data group and the signal characteristic information of each target partial discharge type in the rule table, and determine the feature probability of each waveform characteristic and each analysis characteristic based on the judgment result.

[0117] For each type of target partial discharge, if the waveform or analytical features of the analysis spectrum corresponding to the data set match the corresponding signal feature information, the feature probability corresponding to the waveform or analytical feature is 1; otherwise, the feature probability corresponding to the waveform or analytical feature is 0.

[0118] This application embodiment can determine the type probability Xi according to the eight types of features (i.e., waveform features and analysis features) in the rule table. Among the N real-time analysis results in the data group, if a certain item in the waveform feature and analysis feature matches its corresponding item in the rule table, then Xi=1; otherwise, Xi=0.

[0119] S407: Calculate the type probability P of each data set corresponding to each type of target partial discharge based on the first formula, which is:

[0120] ;

[0121] Where Xi represents the feature probability corresponding to the waveform feature or analysis feature in the data set, q is equal to the total number of waveform features and analysis features in the data set, and 0 < i ≤ q.

[0122] It is understandable that when the data set corresponds to the Mth to the (M+Nth)th acquisition cycle, q = N. The total number of waveform features and analysis features corresponding to data in a single acquisition cycle. Specifically, the total number of waveform features and analysis features corresponding to data in a single acquisition cycle is 8. For example, when N=13, q=13. 8 = 104.

[0123] In this embodiment, steps S405-S407 can be executed after step S403. This allows for the determination of the probability of a partial discharge anomaly existing for each target partial discharge type.

[0124] Furthermore, step S406 may be followed by steps S408-S410.

[0125] S408: If the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to the data set do not match the signal characteristic information corresponding to each type of target partial discharge, determine whether the flight chart characteristics show significant features based on the similarity of flight chart characteristics.

[0126] The similarity of flight chart features varies depending on whether they are significant or insignificant. For example, if the peak-trough analysis shows multiple peaks and the overall trend is increasing, then the flight chart features are significant (or have obvious features), and the similarity is 0.99. If the peak-trough analysis shows multiple peaks or the overall trend is increasing, then the flight chart features are significant, and the similarity is 0.4. If neither condition is met, then the flight chart features are insignificant (or have no obvious features), and the similarity is 0.

[0127] It is worth noting that the value of the flight map feature similarity can be a fixed value set based on operational experience, or a dynamic value that is adaptively adjusted based on actual conditions and needs. For example, the similarity features of each flight map can fluctuate within their corresponding range, and this application embodiment does not specifically limit this.

[0128] Based on this, the embodiments of this application can further determine whether particle discharge exists.

[0129] Understandably, since the eight characteristics of particulate discharge are similar to those of no partial discharge anomalies, the presence of particulate discharge cannot be determined based on these eight characteristics. Therefore, if the current waveform and analysis characteristics do not match the air gap discharge type, surface discharge type, suspension discharge type, and corona discharge type in the rule table, the presence of particulate discharge can be further determined based on the flight chart characteristics.

[0130] S409: If the flight map features are significant, the similarity of the flight map features corresponding to the significant features will be determined as the type probability corresponding to the particle discharge type.

[0131] If the flight map features are significant, then particle discharge is evident, confirming the presence of an anomaly. In this case, the type probability corresponding to the particle discharge type can be taken as the similarity of the flight map features.

[0132] In this way, the probability of particle discharge partial discharge anomaly can be obtained.

[0133] S410: Based on the type probability of each data group corresponding to each type of partial discharge, obtain the partial discharge diagnosis result.

[0134] In this way, the partial discharge diagnosis results of the edge-side devices corresponding to each data group can be obtained. The partial discharge diagnosis results of each edge-side device are used to indicate the probability and type of partial discharge anomaly in the power grid terminal it covers.

[0135] In some implementations, steps S401-S410 can be performed by the data analysis model.

[0136] Furthermore, step S410 may specifically include the following steps S4101-S4103.

[0137] S4101: For each data set, determine the maximum probability value in the type probability corresponding to each partial discharge type, and count the number of type probabilities equal to the maximum probability value.

[0138] Understandably, the specific value of the maximum probability depends on the actual calculation.

[0139] S4102: If the number of type probabilities equal to the maximum probability value is one, determine that the partial discharge type corresponding to the data group is the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the abnormal probability is equal to the maximum probability value, and obtain the partial discharge diagnosis result of the edge-side device corresponding to the data group.

[0140] In other words, in this application embodiment, the partial discharge type with the highest probability among the five types of partial discharge (target partial discharge type and particle discharge type) can be taken as the final analysis result.

[0141] S4103: If there are multiple types of probabilities equal to the maximum probability value, determine that the partial discharge type corresponding to the data group is a random one of the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the abnormal probability is equal to the maximum probability value, and obtain the partial discharge diagnosis result of the edge-side device corresponding to the data group.

[0142] In other words, if multiple partial discharge types have the same type probability, then any one of the partial discharge types is randomly selected as the final analysis result.

[0143] It should be noted that, please refer to [link / reference needed]. Figure 2 Furthermore, embodiments of this application can also utilize data analysis models to perform comprehensive diagnosis of batch sampled data (data sets), i.e., batch sampling diagnosis, to eliminate sporadic risks. Specifically, embodiments of this application can input a data set formed by aggregating multiple consecutive sampling cycles from the same edge-side device into a data analysis model. The data analysis model is a pre-trained model that can perform joint analysis of the mechanistic characteristics of all samples within the data set at once, directly outputting the partial discharge type and anomaly probability of the power grid terminal covered by the edge-side device, thereby avoiding random errors caused by single sampling and achieving stable diagnosis of batch data.

[0144] As can be seen, the embodiments of this application provide a voting mechanism in which each real-time analysis result in the data group can participate in the voting process of the type probability P, so as to obtain the type probability result of multiple collection cycles. The partial discharge type corresponding to the largest type probability is used as the final diagnosis result. At the same time, the largest type probability is directly set as the abnormal probability of partial discharge anomaly in the power grid terminal, so as to realize the one-time output of partial discharge type and risk probability.

[0145] In some implementations, embodiments of this application can analyze the real-time analysis results and target data corresponding to a single acquisition cycle, and determine whether the waveform characteristics or analysis characteristics of the analysis spectrum corresponding to a single acquisition cycle match the signal characteristic information corresponding to the target partial discharge type. If the characteristics match, the probability is 1; if the characteristics do not match, the probability is 0. Then, the type probability P' of the data in a single acquisition cycle corresponding to various target partial discharge types is calculated based on the second formula. The second formula is:

[0146] ;

[0147] Where Xi' represents the feature probability corresponding to the waveform feature or analysis feature in the data of a single acquisition cycle, t is the total number of waveform features and analysis features, and 0 < i ≤ t.

[0148] In this embodiment of the application, t equals 8, i = 0, 1, ..., 7, and the second formula is specifically as follows: .

[0149] And / or, embodiments of this application can calculate the type probability corresponding to a particle discharge type when the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to a single acquisition cycle do not match the signal characteristic information corresponding to each target partial discharge type.

[0150] Then, by selecting the partial discharge type corresponding to the maximum type probability (or maximum probability value) (if there is more than one maximum type probability, one of the target partial discharge type or particle discharge type corresponding to the maximum type probability can be randomly selected) as the partial discharge type corresponding to the acquisition period, and the maximum probability value is determined as the abnormal probability corresponding to the acquisition period.

[0151] This allows for anomaly probability assessment of a particular partial discharge monitoring sample, and further subdivision of the partial discharge type.

[0152] In some implementations, embodiments of this application can aggregate partial discharge types and anomaly probabilities for N acquisition cycles. If the number of times a certain partial discharge type occurs in the N acquisition cycles exceeds a preset threshold, and / or the number of times the anomaly probability exceeds a preset threshold exceeds a preset threshold, the partial discharge type can be used as the final partial discharge analysis result, and the average of all type probabilities corresponding to the partial discharge type can be taken as the corresponding anomaly probability.

[0153] Furthermore, the method provided in this application embodiment may also include the following step S601: aggregating the partial discharge type and anomaly probability corresponding to each edge-side device to form a monitoring list.

[0154] Understandably, after determining the partial discharge type and anomaly probability, the embodiments of this application can summarize and integrate the data. The partial discharge type and anomaly probability of each grid terminal covered by each edge-side device are aggregated to form a detailed monitoring list. This provides maintenance personnel with comprehensive and intuitive information, facilitating the monitoring and management of the operation of the grid terminals covered by each edge-side device.

[0155] For some implementation methods, please refer to [link / reference]. Figure 2 On the user end, maintenance personnel, repair personnel, and managers can all log in to the cloud intelligent agent's web page to view partial discharge monitoring, diagnosis, and early warning data. They can view diagnostic results and handling suggestions, and perform manual review. In this way, users can view the overall situation of partial discharge monitoring of power grid terminals, such as whether there are power grid terminals with a high probability of partial discharge anomalies, whether there are power grid terminals with excessive humidity, and whether there are power grid terminals diagnosed with partial discharge and being handled, thus identifying the stations that require special attention.

[0156] In some implementations, in addition to generating a monitoring list, embodiments of this application can also generate early warnings for power grid terminals with partial discharge anomalies, classify early warning levels according to probability, provide anomaly ranking and handling suggestions, and provide a list of terminals that customers need to pay close attention to.

[0157] For some implementation methods, please refer to [link / reference]. Figure 2 This application embodiment can also perform partial discharge influencing factor analysis, such as analyzing the relationship between partial discharge and temperature and humidity, or partial discharge-temperature influencing factor analysis, which can be performed by calling a data analysis model based on the actual situation. This application embodiment does not make specific limitations in this regard.

[0158] For some implementation methods, please refer to [link / reference]. Figure 2 Furthermore, it can also perform inter-cabinet interference analysis, ledger influencing factor analysis, etc., but this application does not specifically limit these aspects.

[0159] In some implementations, the embodiments of this application may also perform step S602: for each edge-side device, generate disposal recommendations based on partial discharge type, anomaly probability, partial discharge monitoring data, environmental monitoring data and / or power grid terminal ledger data.

[0160] The ledger data records the basic information, operating status, and / or maintenance history of power grid terminals, covering their entire lifecycle. For example, the ledger data may include basic equipment information (such as model), operating parameters, maintenance records, fault records, and environmental monitoring data for the power grid terminals.

[0161] In some implementations, the embodiments of this application may also perform the following steps S603-S604.

[0162] S603: Obtain the decision result for the partial discharge diagnosis result, and form positive samples and negative samples; positive samples correspond to the decision result being consistent with the partial discharge diagnosis result, and negative samples correspond to the decision result being inconsistent with the partial discharge diagnosis result.

[0163] Understandably, see [link / reference] Figure 2 Users can make a judgment on the diagnostic results, that is, determine whether there is partial discharge. If it is a false alarm, that is, there is no partial discharge abnormality, the embodiments of this application can record the false alarm and use its related data as negative samples to train the model. If it is not a false alarm, that is, there is partial discharge abnormality, the diagnostic results can be used as positive samples to train the model. Furthermore, users can perform manual handling actions for partial discharge abnormalities.

[0164] S604: Use positive and negative samples to train a data analysis model.

[0165] By combining system diagnosis with manual diagnosis, continuously accumulating sample data, and through regular automatic learning and training, the accuracy of the partial discharge diagnosis model is improved.

[0166] For further details, please refer to [link / reference]. Figure 2 This application embodiment can determine whether frequency adjustment is necessary, including adjusting the partial discharge monitoring frequency (or adjusting the monitoring frequency) and the signal sampling frequency. For example, when a possible partial discharge anomaly is detected, monitoring can be strengthened, i.e., increasing the signal sampling frequency of the sensing terminal, as well as the monitoring frequency and data reporting frequency of the edge-side device. Alternatively, the normal frequency can be restored. This adjustment action can be notified to the edge-side device by the cloud intelligent agent, and then the edge-side device can issue an instruction to control the sensing terminal to adjust the sampling frequency.

[0167] Figure 4 This is a schematic diagram of the intelligent question-and-answer process provided in an embodiment of this application.

[0168] like Figure 4As shown in the embodiments of this application, the partial discharge monitoring and maintenance method based on intelligent agents can be used to implement an intelligent maintenance assistant, which can specifically include four layers.

[0169] The first layer is the user interaction layer: for operations and maintenance personnel / inspection personnel / management personnel, they can interact with the assistant by logging into the cloud intelligence agent's web page. They can ask questions on this interface and view the answers generated by the system.

[0170] The second level is the system administrator operation level. System administrators can regularly import relevant knowledge or experience documents into the system to continuously update and expand the system's knowledge base, ensuring that the system can provide accurate and up-to-date information.

[0171] Furthermore, after the knowledge base is updated, the system administrator can initiate model training to ensure that the system can understand and process the new knowledge content.

[0172] The third layer is the upper layer of the intelligent agent application, which is used for question transmission and answer display.

[0173] The fourth level is the text big model, which can be used to process and understand large amounts of text data. After receiving a question, the text big model will use the knowledge it has trained to generate an answer.

[0174] Specifically, the large text model is trained based on pre-set training documents, which include partial discharge monitoring data, environmental monitoring data, power grid terminal ledger data, and / or partial discharge knowledge.

[0175] The method provided in this application embodiment may further include the following steps S701-S702.

[0176] S701: In response to a text question submitted by a user through a web page, the text big model is invoked to parse the question and generate an answer.

[0177] S702: Display the answer on the web page.

[0178] This ensures that users can interact with the cloud intelligence agent through an intuitive web interface and receive satisfactory answers.

[0179] Figure 5 This is a schematic diagram of the first structure of the partial discharge monitoring and maintenance system based on intelligent agents provided in the embodiments of this application.

[0180] Figure 6 This is a schematic diagram of a second structure of the partial discharge monitoring and maintenance system based on intelligent agents provided in an embodiment of this application.

[0181] like Figure 5 and Figure 6As shown in the illustration, this application also provides a partial discharge monitoring and maintenance system based on an intelligent agent. This system may include a power grid terminal 100, a sensing terminal 200, an edge-side device 300, and a cloud intelligent agent 400. The power grid terminal 100 can be a switchgear, and multiple power grid terminals 100 can be located in the same substation room. A substation room can be configured with multiple sensing terminals 200. The sensing terminals 200 can be located on the edge side and are considered edge-side devices. The edge-side device 300 can be a substation node used to manage the power grid terminals 100 and their associated sensing terminals in a substation room. The edge-side device 300 can be located on the edge side. Furthermore, the cloud intelligent agent 400 can be located in the cloud. This forms a cloud-edge-device collaborative architecture.

[0182] Furthermore, the cloud layer can include an application layer, a model layer, and a data layer. The application layer can provide partial discharge (PD) monitoring, PD early warning, PD diagnosis, and PD-assisted decision-making. The model layer can provide comprehensive PD diagnosis using CNN classification models for PD spectra, PD mechanism feature analysis models, and PD influencing factor analysis models. It can also provide LLM text-based large-scale models and manual model training. The data layer can provide monitoring data access and remote adjustment command issuance for monitoring cycles, specifically involving PD monitoring data, environmental monitoring data, equipment ledger data, diagnostic decision-making data, and PD operation and maintenance knowledge.

[0183] The power grid terminal 100 is, for example, a 10kV switchgear, incoming line cabinet, feeder cabinet, bus tie cabinet, bus tie isolation cabinet, PT cabinet and / or substation transformer in a distribution station room.

[0184] In some implementations, the sensing terminal 200 may include a partial discharge sensor 201 and a temperature and humidity sensor 202. Each substation room may be equipped with a temperature and humidity sensor 202 to collect environmental monitoring data, such as temperature and humidity. Additionally, each switchgear 101 in each substation room may be equipped with a partial discharge sensor 201, which may include dual ultrasonic partial discharge sensors. Alternatively, the partial discharge sensor 201 may also be a visual ultrasonic monitoring device, which may include temperature, humidity, and ozone sensors, and may capture ultrasonic waves and overlay them with camera images to achieve visualization.

[0185] It is worth noting that the sensing terminal 200 can specifically collect data from the switch cabinets in the substation room, as well as from the incoming line cabinets, feeder cabinets, bus tie cabinets, bus tie isolation cabinets, PT cabinets, and transformers within the substation.

[0186] Furthermore, the edge device 300 is configured to generate target data and report it to the cloud agent 400; the target data is determined based on at least one power grid terminal 100 covered by the edge device 300, and the target data includes monitoring data and pre-diagnostic data, which is generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data;

[0187] The cloud intelligent agent 400 is configured to: periodically collect target data reported by at least one edge device 300; when target data is acquired in each cycle, perform data analysis on the target data using a data analysis model to obtain real-time analysis results; for each edge device 300, aggregate the real-time analysis results and target data within an analysis interval to obtain a data set; wherein, an analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, where M and N are preset values; and analyze the data set using a data analysis model to obtain partial discharge diagnosis results, which include the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0188] In some implementations, the monitoring data includes partial discharge monitoring data; the cloud agent 400 is specifically configured to: plot the partial discharge monitoring data to obtain an analysis spectrum; the analysis spectrum includes a partial discharge pulse signal spectrum; and use a spectrum classification and recognition model to perform waveform analysis on the partial discharge pulse signal spectrum to determine the waveform characteristics of the analysis spectrum; wherein, the waveform characteristics include one or more of waveform type, peak information, and trough information, and the waveform type includes single peak, double peak, multi-peak, and stable peak; the real-time analysis results include waveform characteristics.

[0189] In some implementations, the analysis spectrum also includes a partial discharge phase analysis spectrum. The cloud agent 400 is further configured to: perform feature calculations on the partial discharge pulse signal spectrum using a mechanism model to obtain analysis features, including amplitude dispersion features, pulse equalization features, polarity features, phase clustering features, 50Hz component significance features, 100Hz component significance features, and 50 / 100Hz component numerical relationship features; perform feature calculations on the partial discharge phase analysis spectrum using a mechanism model to obtain flight pattern features; and the real-time analysis results include the analysis features and flight pattern features.

[0190] In some implementations, the pre-diagnostic data includes partial discharge alarm information; the partial discharge type includes target partial discharge type and particulate discharge type; the cloud agent 400 is also configured to: determine whether the number of partial discharge alarm information in the data group is greater than a preset threshold; if the number of partial discharge alarm information in the data group is greater than the preset threshold, obtain a rule table from the database; the rule table includes signal feature information corresponding to different target partial discharge types; the target partial discharge types include, but are not limited to, air gap discharge type, surface discharge type, suspension discharge type, and corona discharge type; according to the rule table, determine the degree of matching between the waveform features and analysis features of the analysis spectrum corresponding to each data group and the signal feature information of each target partial discharge type in the rule table, and determine the feature probability of each waveform feature and each analysis feature based on the judgment result; wherein, for each target partial discharge type, if the waveform features or analysis features of the analysis spectrum corresponding to the data group conform to the corresponding signal feature information, the feature probability corresponding to the waveform feature or analysis feature is 1, and if it does not conform to the signal feature information, the feature probability corresponding to the waveform feature or analysis feature is 0; calculate the type probability P of each data group corresponding to each target partial discharge type based on the first formula, the first formula being:

[0191] ;

[0192] Where Xi represents the feature probability corresponding to the waveform feature or analysis feature in the data set, q is equal to the total number of waveform features and analysis features in the data set, and 0 < i ≤ q;

[0193] If the waveform and analysis features of the analysis spectrum corresponding to the data set do not match the signal feature information corresponding to each target partial discharge type, the flight pattern feature similarity is used to determine whether the flight pattern feature presents a significant feature; if the flight pattern feature presents a significant feature, the similarity of the flight pattern feature corresponding to the significant feature is determined as the type probability corresponding to the particle discharge type; based on the type probability of each data set corresponding to each partial discharge type, the partial discharge diagnosis result is obtained.

[0194] In some implementations, the cloud agent 400 is further configured to: for each data group, determine the maximum probability value among the type probabilities of each partial discharge type corresponding to the data group, and count the number of type probabilities equal to the maximum probability value; if the number of type probabilities equal to the maximum probability value is one, determine that the partial discharge type corresponding to the data group is the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the anomaly probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge device corresponding to the data group; if the number of type probabilities equal to the maximum probability value is multiple, determine that the partial discharge type corresponding to the data group is one of the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the anomaly probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge device corresponding to the data group.

[0195] In some implementations, the cloud agent 400 is also configured to: aggregate the partial discharge type and anomaly probability corresponding to each edge device 300 to form a monitoring list; and / or, for each edge device 300, generate disposal suggestions based on the partial discharge type, anomaly probability, partial discharge monitoring data, environmental monitoring data, and / or the ledger data of the power grid terminal 100, wherein the ledger data records the basic information, operating status, and / or maintenance history of the power grid terminal; and / or, obtain the judgment result for the partial discharge diagnosis result to form positive samples and negative samples; positive samples correspond to the judgment result being consistent with the partial discharge diagnosis result, and negative samples correspond to the judgment result being inconsistent with the partial discharge diagnosis result; and use the positive samples and negative samples to train the data analysis model.

[0196] In some implementations, the cloud agent 400 is also configured to: respond to a text question raised by a user through a web page, invoke a large text model to parse the question and generate an answer; wherein, the large text model is trained based on a preset training document, which includes partial discharge monitoring data, environmental monitoring data, ledger data of the power grid terminal 100 and / or partial discharge knowledge; and display the answer on the web page.

[0197] In some implementations, the pre-diagnostic data includes partial discharge alarm information. The cloud intelligent agent 400 is also configured to: for each edge device 300, determine whether there is a discharge interference signal in the power grid terminal 100 it covers based on the data group; if there is no discharge interference signal, determine that the partial discharge type is no abnormality and the probability of abnormality is equal to 0; if there is a discharge interference signal, determine whether the number of partial discharge alarm information in the data group is greater than a preset threshold; if the number of partial discharge alarm information in the data group is not greater than the preset threshold, determine that the partial discharge type is no abnormality and the probability of abnormality is equal to 0.

[0198] In summary, the partial discharge monitoring and maintenance method and system based on intelligent agents provided in this application embodiment consists of the following three parts.

[0199] (1) Sensing terminal 200 level (end): Using sensors to realize local real-time monitoring of partial discharge signals of power grid terminal 100 (e.g., switch cabinet 101); and by importing the live detection work report into the operation and maintenance system, to realize unified management of on-site online monitoring and live detection data.

[0200] (2) Edge computing layer (edge): Based on edge side device 300 (e.g., station node), realize centralized access to sensing terminal 200 (e.g., various monitoring devices in power distribution room), and realize the overall supervision and time synchronization of local sensing terminal 200 with a certain computing power, and can perform partial discharge pre-diagnosis.

[0201] (3) Cloud platform level (cloud): The cloud intelligent agent 400 relies on four major categories of data, including partial discharge online monitoring data of power grid terminal 100, temperature and humidity monitoring data, power grid terminal 100 (e.g. switch cabinet 101) ledger data, partial discharge knowledge and expert experience, to construct two major models: data analysis model and text big model, so as to realize end-to-end application of partial discharge monitoring, diagnosis, early warning and disposal.

[0202] The partial discharge monitoring and maintenance method and system based on intelligent agents provided in this application can realize real-time monitoring and fault early warning of indoor electrical equipment in substations through cloud-edge-device collaboration, which helps to improve the operational reliability and maintenance efficiency of electrical equipment.

[0203] Figure 7 This is a schematic diagram of the structure of the partial discharge monitoring and maintenance device based on intelligent agents provided in the embodiments of this application.

[0204] like Figure 7 As shown in the figure, this application provides a partial discharge monitoring and maintenance device based on an intelligent agent, applied to a cloud intelligent agent. The device includes:

[0205] The acquisition module 1001 is configured to periodically acquire target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device. The target data includes monitoring data and pre-diagnostic data. The pre-diagnostic data is generated when partial discharge signals and / or discharge interference signals are identified in the monitoring data.

[0206] The first analysis module 1002 is configured to: when the target data is acquired in each cycle, use the data analysis model to perform data analysis on the target data and obtain real-time analysis results;

[0207] The aggregation module 1003 is configured to aggregate real-time analysis results and target data within an analysis interval for each edge device to obtain a data set; wherein, an analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, and M and N are preset values;

[0208] The second analysis module 1004 is configured to analyze the data set using a data analysis model to obtain partial discharge diagnosis results, which include the probability of partial discharge anomalies in the power grid terminal and the type of partial discharge.

[0209] In a specific implementation, the present invention also provides a computer storage medium, wherein the computer storage medium may store a program, and when the program is executed, it may include some or all of the steps of the various embodiments of the agent-based partial discharge monitoring and maintenance method provided by the present invention. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0210] It is readily understood that, based on the several embodiments provided in this application, those skilled in the art can combine, split, or reorganize the embodiments of this application to obtain other embodiments, none of which exceed the protection scope of this application.

[0211] The above detailed embodiments further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A partial discharge monitoring and maintenance method based on intelligent agents, characterized in that, Applied to cloud intelligent agents, the method includes: Periodically collect target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device. The target data includes monitoring data and pre-diagnostic data. The pre-diagnostic data is generated when a partial discharge signal is identified in the monitoring data. The pre-diagnostic data includes partial discharge alarm information. When the target data is acquired in each cycle, the target data is analyzed using a data analysis model to obtain real-time analysis results; the real-time analysis results include waveform features, analysis features, and flight chart features. For each edge device, the real-time analysis results and the target data within an analysis interval are aggregated to obtain a data set; wherein, the analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, and M and N are preset values; The data set is analyzed using the data analysis model to obtain partial discharge diagnosis results. The partial discharge diagnosis results include the probability of partial discharge anomaly in the power grid terminal and the type of partial discharge. The partial discharge type includes target partial discharge type and particulate discharge type. The step of using a data analysis model to analyze the data set and obtain partial-flood diagnostic results includes: Determine whether the number of partial discharge alarm messages in the data group is greater than a preset threshold; If the number of partial discharge alarm messages in the data group is greater than the preset threshold, a rule table is retrieved from the database; the rule table includes signal characteristic information corresponding to different target partial discharge types. According to the rule table, the degree of matching between the waveform features and analytical features of the analysis spectrum corresponding to each data group and the signal feature information of each target partial discharge type in the rule table is determined, and the feature probability of each waveform feature and each analytical feature is determined based on the judgment result; wherein, for each target partial discharge type, if the waveform features or analytical features of the analysis spectrum corresponding to the data group match the corresponding signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 1, and if it does not match the signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 0. For each set of data, the average value of the target feature probabilities corresponding to a certain type of target partial discharge is summed to obtain the type probability of the data set corresponding to that type of target partial discharge, and then the type probability of the data set corresponding to each type of target partial discharge is obtained; the target feature probability includes the feature probability corresponding to the waveform feature or analysis feature in the data set; If the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to the data group do not match the signal characteristic information corresponding to each of the target partial discharge types, the flight pattern characteristics are judged to be significant based on the similarity of the flight pattern characteristics. If the flight map features exhibit significant characteristics, the similarity of the flight map features corresponding to the significant characteristics will be determined as the type probability corresponding to the particle discharge type. The partial discharge diagnosis result is obtained based on the type probability of each data set corresponding to each type of partial discharge.

2. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 1, characterized in that, The monitoring data includes partial discharge monitoring data; the step of using a data analysis model to analyze the target data and obtain real-time analysis results includes: The partial discharge monitoring data is plotted to obtain an analytical spectrum; the analytical spectrum includes a partial discharge pulse signal spectrum. The partial discharge pulse signal spectrum is analyzed using a spectrum classification and recognition model to determine the waveform characteristics of the analyzed spectrum; wherein, the waveform characteristics include one or more of waveform type, peak information, and trough information, and the waveform type includes single peak, double peak, multi-peak, and stable peak.

3. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 2, characterized in that, The analysis spectrum also includes a partial discharge phase analysis spectrum. After the step of plotting the partial discharge monitoring data to obtain the analysis spectrum, the method further includes: The partial discharge pulse signal spectrum is analyzed using a mechanistic model to obtain analytical features, including amplitude dispersion features, pulse uniformity features, polarity features, phase clustering features, 50Hz component significance features, 100Hz component significance features, and 50 / 100Hz component numerical relationship features. The characteristics of the partial discharge phase analysis spectrum are calculated using the aforementioned mechanism model to obtain flight pattern features.

4. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 3, characterized in that, The target partial discharge types include air gap discharge type, surface discharge type, suspended discharge type, and corona discharge type; the step of summing the average value of the target feature probabilities corresponding to a target partial discharge type for each data set to obtain the type probability of the data set corresponding to that target partial discharge type, and then obtaining the type probability of the data set corresponding to each target partial discharge type, includes: The type probability P corresponding to each target localization type for each data set is calculated based on the first formula, which is: ； Where Xi represents the feature probability corresponding to the waveform feature or analysis feature in the data set, q is equal to the total number of waveform features and analysis features in the data set, and 0 < i ≤ q.

5. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 4, characterized in that, The step of obtaining the partial discharge diagnosis result based on the type probability of each data group corresponding to each partial discharge type includes: For each data set, determine the maximum probability value of the type probability corresponding to each local discharge type, and count the number of type probabilities equal to the maximum probability value; If the number of type probabilities equal to the maximum probability value is one, determine that the partial discharge type corresponding to the data group is the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the abnormal probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge-side device corresponding to the data group; If there are multiple types of probabilities equal to the maximum probability value, determine that the partial discharge type corresponding to the data group is a random one of the target partial discharge type or particle discharge type corresponding to the maximum probability value, and determine that the abnormal probability is equal to the maximum probability value, thereby obtaining the partial discharge diagnosis result of the edge-side device corresponding to the data group.

6. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 2, characterized in that, The monitoring data also includes environmental monitoring data; the method further includes: The partial discharge type and anomaly probability of each edge device are aggregated to form a monitoring list; And / or, For each of the aforementioned edge-side devices, a handling suggestion is generated based on the partial discharge type, the anomaly probability, the partial discharge monitoring data, the environmental monitoring data, and / or the ledger data of the power grid terminal. The ledger data is data that records the basic information, operating status, and / or maintenance history of the power grid terminal. And / or, Obtain a decision result for the partial discharge diagnosis result, forming positive samples and negative samples; the positive sample corresponds to the decision result being consistent with the partial discharge diagnosis result, and the negative sample corresponds to the decision result being inconsistent with the partial discharge diagnosis result; The data analysis model is trained using the positive and negative samples.

7. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 6, characterized in that, The method further includes: In response to a text question submitted by a user through a web page, a large text model is invoked to parse the question and generate an answer; wherein, the large text model is trained based on a preset training document, which includes the partial discharge monitoring data, the environmental monitoring data, the ledger data of the power grid terminal, and / or partial discharge knowledge; The answer will be displayed on the web page.

8. The partial discharge monitoring and maintenance method based on intelligent agents according to claim 1, characterized in that, The pre-diagnostic data includes partial discharge alarm information; The step of analyzing the data set using the data analysis model to obtain the partial-exposure diagnostic results further includes: For each of the aforementioned edge-side devices, the presence of discharge interference signals in the power grid terminals covered by the data set is determined. If the discharge interference signal is not present, the partial discharge type is determined to be without anomaly, and the anomaly probability is determined to be equal to 0. If the discharge interference signal is present, determine whether the number of partial discharge alarm messages in the data group is greater than a preset threshold; If the number of partial discharge alarm messages in the data group is not greater than the preset threshold, the partial discharge type is determined to be without anomalies, and the anomaly probability is determined to be equal to 0.

9. A partial discharge monitoring and maintenance system based on intelligent agents, characterized in that, The system includes: edge devices and cloud intelligent agents; The edge device is configured to generate target data and report it to the cloud agent; the target data is determined based on at least one power grid terminal covered by the edge device, and the target data includes monitoring data and pre-diagnostic data, wherein the pre-diagnostic data is generated when a partial discharge signal is identified in the monitoring data; The cloud agent is configured as follows: Periodically collect target data reported by at least one edge-side device; the pre-diagnostic data includes partial discharge alarm information; When the target data is acquired in each cycle, the target data is analyzed using a data analysis model to obtain real-time analysis results; the real-time analysis results include waveform features, analysis features, and flight chart features. For each edge device, the real-time analysis results and the target data within an analysis interval are aggregated to obtain a data set; wherein, the analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, and M and N are preset values; The data set is analyzed using the data analysis model to obtain partial discharge diagnosis results. The partial discharge diagnosis results include the probability of partial discharge anomaly in the power grid terminal and the type of partial discharge. The partial discharge type includes target partial discharge type and particulate discharge type. The step of using a data analysis model to analyze the data set and obtain partial-flood diagnostic results includes: Determine whether the number of partial discharge alarm messages in the data group is greater than a preset threshold; If the number of partial discharge alarm messages in the data group is greater than the preset threshold, a rule table is retrieved from the database; the rule table includes signal characteristic information corresponding to different target partial discharge types. According to the rule table, the degree of matching between the waveform features and analytical features of the analysis spectrum corresponding to each data group and the signal feature information of each target partial discharge type in the rule table is determined, and the feature probability of each waveform feature and each analytical feature is determined based on the judgment result; wherein, for each target partial discharge type, if the waveform features or analytical features of the analysis spectrum corresponding to the data group match the corresponding signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 1, and if it does not match the signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 0. For each set of data, the average value of the target feature probabilities corresponding to a certain type of target partial discharge is summed to obtain the type probability of the data set corresponding to that type of target partial discharge, and then the type probability of the data set corresponding to each type of target partial discharge is obtained; the target feature probability includes the feature probability corresponding to the waveform feature or analysis feature in the data set; If the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to the data group do not match the signal characteristic information corresponding to each of the target partial discharge types, the flight pattern characteristics are judged to be significant based on the similarity of the flight pattern characteristics. If the flight map features exhibit significant characteristics, the similarity of the flight map features corresponding to the significant characteristics will be determined as the type probability corresponding to the particle discharge type. The partial discharge diagnosis result is obtained based on the type probability of each data set corresponding to each type of partial discharge.

10. A partial discharge monitoring and maintenance device based on intelligent agents, characterized in that, The device, applied to cloud intelligent agents, includes: The acquisition module is configured to periodically acquire target data reported by at least one edge-side device. The target data is determined based on at least one power grid terminal covered by the edge-side device. The target data includes monitoring data and pre-diagnostic data. The pre-diagnostic data is generated when a partial discharge signal is identified in the monitoring data. The pre-diagnostic data includes partial discharge alarm information. The first analysis module is configured to: when the target data is acquired in each cycle, use a data analysis model to perform data analysis on the target data to obtain real-time analysis results; the real-time analysis results include waveform features, analysis features, and flight chart features. The aggregation module is configured to: for each edge device, aggregate the real-time analysis results and the target data within an analysis interval to obtain a data group; wherein, the analysis interval refers to the interval corresponding to the Mth acquisition cycle to the M+Nth acquisition cycle, and M and N are preset values; The second analysis module is configured to: analyze the data set using the data analysis model to obtain partial discharge diagnosis results, wherein the partial discharge diagnosis results include the probability of partial discharge anomaly in the power grid terminal and the partial discharge type; the partial discharge type includes target partial discharge type and particulate discharge type; The second analysis module is specifically configured as follows: Determine whether the number of partial discharge alarm messages in the data group is greater than a preset threshold; If the number of partial discharge alarm messages in the data group is greater than the preset threshold, a rule table is retrieved from the database; the rule table includes signal characteristic information corresponding to different target partial discharge types. According to the rule table, the degree of matching between the waveform features and analytical features of the analysis spectrum corresponding to each data group and the signal feature information of each target partial discharge type in the rule table is determined, and the feature probability of each waveform feature and each analytical feature is determined based on the judgment result; wherein, for each target partial discharge type, if the waveform features or analytical features of the analysis spectrum corresponding to the data group match the corresponding signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 1, and if it does not match the signal feature information, the feature probability corresponding to the waveform feature or analytical feature is 0. For each set of data, the average value of the target feature probabilities corresponding to a certain type of target partial discharge is summed to obtain the type probability of the data set corresponding to that type of target partial discharge, and then the type probability of the data set corresponding to each type of target partial discharge is obtained; the target feature probability includes the feature probability corresponding to the waveform feature or analysis feature in the data set; If the waveform characteristics and analysis characteristics of the analysis spectrum corresponding to the data group do not match the signal characteristic information corresponding to each of the target partial discharge types, the flight pattern characteristics are judged to be significant based on the similarity of the flight pattern characteristics. If the flight map features exhibit significant characteristics, the similarity of the flight map features corresponding to the significant characteristics will be determined as the type probability corresponding to the particle discharge type. The partial discharge diagnosis result is obtained based on the type probability of each data set corresponding to each type of partial discharge.

Images