Intelligent inspection and state evaluation diagnosis method for main transformer

Abstract

This invention relates to the field of power equipment diagnostics, specifically to an intelligent inspection and condition assessment diagnostic method for main transformers. The method includes: first, determining a baseline operating condition based on multi-source online monitoring data and operating conditions; then, calculating operating condition deviations according to preset component domains to determine the intensity of component anomalies and generate abnormal events; further, associating abnormal events based on a component-level fault propagation graph to reconstruct the fault propagation chain and determine candidate fault propagation chains; subsequently, generating inspection tasks based on the discriminative power of the candidate fault propagation chains, updating the candidate fault propagation chains using newly added monitoring data, determining the target fault propagation chain, and outputting condition assessment and diagnostic results. This invention can improve the targeting of anomaly identification and the accuracy of fault judgment.

Description

Technical Field

[0001] This invention relates to the field of power equipment diagnostics, specifically to a method for intelligent inspection and condition assessment diagnostics of main transformers. Background Technology

[0002] The main transformer is a critical piece of equipment in the power grid transmission and transformation chain, and its operating status directly affects the continuity of power supply, equipment lifespan, and operation and maintenance costs. If anomalies are not identified in a timely manner or faults are not located accurately, it can easily lead to the expansion of defects, unplanned outages, and misallocation of maintenance resources. Existing condition monitoring methods mostly rely on single threshold judgments or static result comparisons, which are difficult to reliably complete condition assessments and fault determinations under conditions of changing operating conditions, abnormal concurrency, and insufficient evidence. Summary of the Invention

[0003] This invention provides an intelligent inspection and condition assessment diagnostic method for main transformers, which at least solves the problem of how to accurately identify anomalies, determine fault propagation relationships, and complete closed-loop diagnostic updates under conditions of changing operating conditions and concurrent multi-source anomalies in main transformers.

[0004] This invention provides an intelligent inspection and condition assessment diagnostic method for a main transformer, the method comprising: The operating condition baseline is determined based on the multi-source online monitoring data and operating conditions of the main transformer. The operating condition deviation is calculated based on the multi-source online monitoring data of the main transformer according to the preset component domain, the abnormal intensity of the component is determined based on the operating condition deviation, and abnormal events are generated according to the changes in the abnormal intensity of the component. Based on the component-level fault propagation graph, abnormal events are associated to reconstruct the fault propagation chain, and candidate fault propagation chains are determined based on the consistency of evidence. Inspection tasks are generated based on the discriminative power of candidate fault propagation chains. New monitoring data is acquired based on the inspection tasks, and candidate fault propagation chains are updated based on the new monitoring data. The target fault propagation chain is determined, and the status assessment results and diagnostic results are output.

[0005] In one possible implementation, the operating condition benchmark is determined based on the multi-source online monitoring data and operating conditions of the main transformer, including: performing time synchronization processing and data reliability assessment on the multi-source online monitoring data of the main transformer; dividing the operating process into operating conditions according to load level, cooling status, tap position and ambient temperature; and determining the operating condition benchmark corresponding to each operating condition based on historical healthy operating data.

[0006] In one possible implementation, the operating condition deviation of the multi-source online monitoring data of the main transformer is calculated according to the preset component domains, including: extracting the monitoring features corresponding to each preset component domain; determining the reference value of each monitoring feature under the current operating condition based on the operating condition benchmark; calculating the operating condition deviation corresponding to each preset component domain based on the observed value and reference value of each monitoring feature, and normalizing the operating condition deviation; wherein, the preset component domains include the winding domain, magnetic circuit domain, cooling domain, bushing and lead domain, and insulation medium domain.

[0007] In one possible implementation, the abnormal intensity of a component is determined based on the operating condition deviation, and an abnormal event is generated according to the change in the abnormal intensity of the component. This includes: determining the initial abnormal intensity of the component based on the operating condition deviation and data reliability assessment results corresponding to each monitoring feature in each preset component domain; performing correlation correction on the initial abnormal intensity of the component based on the temporal and spatial proximity relationships of each monitoring feature within the same preset component domain to obtain the abnormal intensity of the component; and generating an abnormal event when the abnormal intensity of the component continuously exceeds a preset threshold and the trend of change changes; wherein, the abnormal event is used to characterize the start time, duration, intensity, and spatial location of the corresponding preset component domain.

[0008] In one possible implementation, the component-level fault propagation graph includes source component nodes, fault propagation edges, and fault stage nodes; fault propagation edges are used to represent the direction of fault propagation between different source component nodes; fault stage nodes are used to represent the evolution stage of the fault propagation chain; and fault propagation edges are associated with time constraints and space constraints.

[0009] In one possible implementation, abnormal events are associated with a component-level fault propagation graph, fault propagation chains are reconstructed, and candidate fault propagation chains are determined based on evidence consistency. This includes: matching abnormal events to source component nodes and fault propagation edges in the component-level fault propagation graph; determining evaluation results characterizing evidence consistency based on the correspondence between the number, occurrence order, and spatial location of abnormal events and source component nodes, fault propagation edges, and fault stage nodes; determining candidate fault propagation chains based on the evaluation results; and determining the evolution stages of the candidate fault propagation chains.

[0010] In one possible implementation, an inspection task is generated based on the distinguishability of the candidate fault propagation chains, including: identifying at least two candidate fault propagation chains with the highest evaluation results; determining the target inspection area and target inspection method based on the difference evidence used to distinguish the at least two candidate fault propagation chains; and generating the inspection task based on the target inspection area and target inspection method.

[0011] In one possible implementation, the target inspection method includes at least one of directional infrared rescanning, partial discharge encrypted sampling, cooling system status verification, and accessory operation status verification; acquiring new monitoring data based on the inspection task, and updating the candidate fault propagation chain based on the new monitoring data, including: acquiring new monitoring data corresponding to the target inspection area; updating the operating condition deviation and component anomaly intensity of the corresponding preset component domain based on the new monitoring data; regenerating the abnormal event based on the updated component anomaly intensity, and updating the candidate fault propagation chain based on the regenerated abnormal event.

[0012] In one possible implementation, if the difference between the evaluation results of the candidate fault propagation chain with the highest evaluation result and the candidate fault propagation chain with the second highest evaluation result is less than a preset threshold, an inspection task is generated; if the difference between the evaluation results is greater than or equal to the preset threshold, the candidate fault propagation chain with the highest evaluation result is determined as the target fault propagation chain.

[0013] In one possible implementation, the state assessment results include at least two of the following: current severity, evolution rate, and diagnostic uncertainty. The diagnostic results include the source component, the fault propagation path, and the evolution stage corresponding to the target fault propagation chain.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By using operating condition benchmark construction technology, the impact of changes in operating conditions was separated, reducing the interference of operating condition fluctuations on anomaly identification; by using component anomaly intensity and anomaly event generation technology, multi-source monitoring information was transformed into traceable anomaly evidence; by using fault propagation chain reconstruction technology, the correlation judgment from the component source to the propagation path was realized; and by using discrimination-driven inspection and update technology, closed-loop convergence of diagnostic results and targeted review were achieved. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a comparison chart of the baseline operating conditions and the measured temperatures in a specific embodiment of the present invention; Figure 3 This is a timing diagram of component abnormal events in a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the infrared temperature rise distribution in the target inspection area in a specific embodiment of the present invention; Figure 5 This is a comparison chart of the evaluation results of candidate fault propagation chains before and after inspection in a specific embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0019] Intelligent inspection and condition assessment diagnosis refers to an integrated technical process encompassing multi-source sensing, condition identification, anomaly correlation, result judgment, and inspection feedback during equipment operation. This process goes beyond simply reading or issuing alarms for individual monitoring parameters; it comprehensively analyzes monitoring information from different operating conditions to continuously assess the equipment's current state, anomaly evolution trends, and potential fault paths, thereby generating targeted inspection and review actions. Based on this, this invention proposes an intelligent inspection and condition assessment diagnosis method for main transformers, enabling dynamic identification of operating states, correlation judgment of anomaly propagation relationships, and closed-loop updates of diagnostic results.

[0020] like Figure 1 As shown, an intelligent inspection and condition assessment diagnostic method for a main transformer includes: The operating condition baseline is determined based on the multi-source online monitoring data and operating conditions of the main transformer. In this embodiment, multi-source online monitoring data and operating condition information of the main transformer are first acquired, and the alignment of various monitoring data is completed according to a unified time reference. Then, the data reliability of various monitoring data is evaluated by combining sampling continuity, data jump situation and cross-channel correspondence. On this basis, the operating process is segmented based on load level, cooling status, tap position and ambient temperature as the basis for dividing the operating conditions. Then, healthy operating data that are consistent with the current operating conditions and have no fault records are selected from historical operating data to form the operating condition benchmark for the corresponding operating conditions. Finally, the operating condition benchmark is output for subsequent operating condition deviation calculation and anomaly identification.

[0021] The operating condition baseline is determined based on the multi-source online monitoring data and operating conditions of the main transformer, including: time synchronization processing and data reliability assessment of the multi-source online monitoring data of the main transformer; dividing the operating process into operating conditions according to load level, cooling status, tap position and ambient temperature; and determining the operating condition baseline corresponding to each operating condition based on historical healthy operation data.

[0022] In one embodiment, to ensure that the operating condition reference accurately reflects the normal response pattern of the main transformer under different operating conditions, further limitations are imposed on time synchronization processing, data reliability assessment, operating condition classification, and operating condition reference generation. Multi-source online monitoring data may include dissolved gas data in oil, load current data, top oil temperature data, winding hot spot temperature estimation data, cooler switching status data, tap position data, core grounding current data, bushing dielectric loss data, infrared thermography data, and ambient temperature data. Since different data sources have different sampling periods, a unified reference time sequence is first selected, then high-frequency sampling data is aggregated according to time windows, and low-frequency sampling data is aligned according to the most recent valid sampling time. When a certain type of data does not have a valid sampling value within the current time window, the data missing marker for that time window is retained, and extrapolated values ​​are not directly used to fill the gap, thus avoiding the introduction of inaccurate operating condition responses.

[0023] Data reliability assessment can be conducted from four aspects: first, sampling integrity, used to determine whether there are significant missing measurements within a continuous time window; second, numerical stability, used to determine whether there are sudden jumps without physical basis or long periods of inactivity; third, channel consistency, used to determine whether there are significant deviations between similar channels or channels with corresponding relationships; and fourth, state correspondence, used to determine whether the monitoring data matches the actual operating conditions. For example, if the load current rises significantly while the temperature and cooling status remain completely unchanged for a long period, or if the cooler is in operation but the heat dissipation monitoring data shows no response, such data can be judged as low-reliability data. The data reliability assessment results can be given in a graded manner, divided into three levels: high reliability, medium reliability, and low reliability. The grading thresholds are set based on the statistical fluctuation range in historical healthy operation data, or they can be set according to the monitoring stability requirements in the on-site operation procedures.

[0024] When dividing operating conditions, segments are not directly cut according to fixed durations, but rather based on changes in load level, cooling status switching, tap position changes, and ambient temperature range variations. When any key operating condition quantity exceeds a preset change threshold, a new operating condition segment is entered. This change threshold can be set based on common fluctuation ranges in historical operating data, and is primarily used to distinguish between normal fluctuations and operating condition transitions. After the operating condition segment is determined, healthy operating data with consistent operating condition labels, complete operating records, and no corresponding alarms, maintenance, or fault handling events are selected from historical operating data as reference samples for that segment. Subsequently, based on the reference samples, the normal value ranges, trends, and channel correspondences of various monitoring characteristics under the corresponding operating conditions are extracted to form the operating condition baseline. The operating condition baseline can include reference values, allowable fluctuation ranges, and normal transition characteristics during operating condition transitions. If a certain operating condition lacks sufficient historical healthy operating data, healthy operating data from adjacent operating conditions are used for smoothing supplementation, and the source of the supplementation is recorded to avoid the current operating condition lacking a reference basis in subsequent deviation calculations. In this way, the operating condition baseline is not a fixed set of thresholds, but a dynamic reference corresponding to the current load level, cooling status, tap position and ambient temperature, which can be directly used to identify the degree to which the current operating state deviates from the normal response.

[0025] The operating condition deviation is calculated based on the multi-source online monitoring data of the main transformer according to the preset component domain, the abnormal intensity of the component is determined based on the operating condition deviation, and abnormal events are generated according to the changes in the abnormal intensity of the component. After obtaining the operating condition baseline, the operating condition deviation of the multi-source online monitoring data of the main transformer is calculated according to the preset component domains. First, various monitoring data at the current moment are mapped to the corresponding preset component domains. Then, based on the operating condition baseline, reference values ​​for each monitoring feature under the current operating conditions are determined. The difference between the observed values ​​and the reference values ​​is compared to obtain the operating condition deviation for each preset component domain. Based on this, the initial abnormal state of each preset component domain is determined by combining the data reliability assessment results. Then, based on whether different monitoring features within the same preset component domain are continuous in time and adjacent in space, the initial abnormal state is correlated and corrected to form the component abnormality intensity. When the component abnormality intensity continuously exceeds the abnormality judgment threshold, and the trend of change shows a switch from stable to increasing, from increasing to plateau, or from plateau to decay, an abnormal event is generated. The abnormal event is used to record the abnormality start point, duration interval, abnormality strength, and abnormality location of the corresponding preset component domain, providing input for subsequent fault propagation chain reconstruction.

[0026] The operating condition deviation is calculated based on the multi-source online monitoring data of the main transformer according to the preset component domains. This includes: extracting the monitoring features corresponding to each preset component domain; determining the reference value of each monitoring feature under the current operating condition based on the operating condition benchmark; calculating the operating condition deviation corresponding to each preset component domain based on the observed value and reference value of each monitoring feature, and normalizing the operating condition deviation. The preset component domains include the winding domain, magnetic circuit domain, cooling domain, bushing and lead domain, and insulation medium domain.

[0027] In one embodiment, the operating condition deviation calculation revolves around preset component domains. These preset component domains include a winding domain, a magnetic circuit domain, a cooling domain, a bushing and lead domain, and an insulation medium domain. Different preset component domains correspond to different sets of monitoring features. The winding domain may use load current, estimated winding hot spot temperature, top oil temperature change rate, and changes in oil gas related to winding heating; the magnetic circuit domain may use core grounding current, temperature rise in the fixed magnetic circuit area, and local temperature distribution related to magnetic flux anomalies; the cooling domain may use cooler switching status, cooler inlet and outlet temperature difference, oil pump operating status, and top oil temperature drop rate; the bushing and lead domain may use bushing dielectric loss, bushing capacitance, lead connection temperature, and partial discharge monitoring results; and the insulation medium domain may use carbon monoxide, carbon dioxide, and temperature rise characteristics related to insulation aging.

[0028] After the monitoring features are extracted, the corresponding reference values ​​are found in the operating condition baseline based on the current operating conditions. The reference values ​​are not single, fixed numbers, but rather baseline ranges or curves corresponding to load levels, cooling status, tap positions, and ambient temperatures. When the current operating condition is between two adjacent operating conditions, a transitional reference value can be determined from the two operating condition baselines according to the distance principle to reduce abrupt changes at the operating condition switching boundary. After comparing the observed values ​​with the reference values, the deviation of each monitoring feature is obtained. Since different monitoring features have different dimensions and fluctuation amplitudes, the deviations also need to be normalized. The normalization process preferably uses the allowable fluctuation range under the current operating conditions as a unified scale, allowing high-fluctuation features and low-fluctuation features to be compared within the same judgment framework.

[0029] If a monitoring feature is missing within the current time window, or if the data reliability assessment indicates that the monitoring feature is in a low-reliability state, it will not be directly deleted. Instead, it will be retained as a low-reliability input to reduce its impact during subsequent anomaly intensity calculations. This is because some sensing channels may experience short-term distortion or sampling interruptions during main transformer operation. Simply removing these might lead to incomplete anomaly representations within the same preset component domain. After the operating condition deviation calculation is completed, each preset component domain generates a set of deviation results corresponding to the current operating condition. These deviation results retain both the deviation degree of individual monitoring features and the correspondence between monitoring features within the same preset component domain, which can then be directly used to form component anomaly intensity and identify abnormal events.

[0030] Changes in abnormal intensity generate abnormal events, including: determining the initial component abnormal intensity based on the operating condition deviation and data reliability assessment results corresponding to each monitoring feature in each preset component domain; correlating and correcting the initial component abnormal intensity based on the temporal and spatial proximity relationships of each monitoring feature within the same preset component domain to obtain the component abnormal intensity; generating an abnormal event when the component abnormal intensity continuously exceeds a preset threshold and the trend of change changes; wherein, the abnormal event is used to characterize the start time, duration, intensity, and spatial location of the corresponding preset component domain.

[0031] In one embodiment, the generation of abnormal events revolves around the formation and identification of changes in component abnormality intensity. Initial component abnormality intensity is determined first based on the operating condition deviations and data reliability assessment results corresponding to monitoring characteristics within each preset component domain. Specifically, this can be calculated comprehensively according to the importance, reliability level, and deviation magnitude of the monitoring characteristics. Monitoring characteristics that directly reflect local anomalies, such as local temperature rise, enhanced partial discharge, and abnormal core grounding current, can be assigned higher base weights; while auxiliary characterization monitoring characteristics, such as indirect fluctuations caused by changes in ambient temperature, can be assigned lower base weights.

[0032] The data reliability assessment results are used to adjust the contribution of each monitoring feature to the initial component anomaly intensity. When the data reliability is high, the original contribution is maintained; when the data reliability is medium, the contribution is reduced proportionally; and when the data reliability is low, it only serves as an anomaly indication and is not used as the primary criterion. After the initial component anomaly intensity is formed, it is further correlated and corrected based on the temporal and spatial proximity relationships of each monitoring feature within the same preset component domain. Temporal proximity is used to determine whether multiple anomalies occur consecutively within a continuous time window. If multiple anomalies within the same preset component domain occur consecutively at similar times, it indicates the existence of a persistent anomaly within that preset component domain, and the component anomaly intensity can be increased; if the time interval between multiple anomalies is too large, no enhancement is performed. Spatial proximity is used to determine whether multiple anomalies are concentrated in adjacent locations on the main transformer structure. Spatial locations can be identified using infrared temperature measurement area numbers, bushing location numbers, cooler group numbers, lead connection part numbers, etc.

[0033] If multiple anomalies are concentrated in the same or adjacent areas, it indicates spatial clustering of anomalies, and the component anomaly intensity can be further increased. After correlation correction, the final component anomaly intensity is obtained. The generation conditions for anomaly events include two aspects: first, the component anomaly intensity continuously exceeds the anomaly judgment threshold; second, the trend of component anomaly intensity changes. The anomaly judgment threshold can be set based on the maximum normal fluctuation level in historical healthy operation data, or it can be set to a fixed value higher than the upper limit of normal fluctuation based on on-site operation and maintenance experience. The trend change is used to distinguish between short-term disturbances and real anomaly processes. The trend can be divided into three categories: growth, stability, and decay. When the component anomaly intensity changes from stable to growth, from growth to stability, or from stability to decay, an anomaly stage change can be determined. After generating an anomaly event, the start time, end time, duration, anomaly intensity, and spatial location of the anomaly event need to be recorded. The anomaly intensity can be represented by the combined peak intensity and average intensity during the event period, and the spatial location can be represented by the specific equipment area within the corresponding preset component domain. In this way, abnormal events can reflect whether a certain preset component domain is abnormal, as well as when the abnormality occurs, how long it lasts, and where it is mainly concentrated, thus providing a direct basis for subsequent fault propagation relationship judgment.

[0034] Based on the component-level fault propagation graph, abnormal events are associated to reconstruct the fault propagation chain, and candidate fault propagation chains are determined based on the consistency of evidence. After obtaining abnormal events, a component-level fault propagation graph is constructed, and correlation analysis is performed on the abnormal events based on the component-level fault propagation graph. First, the abnormal events are mapped to the source component nodes according to the corresponding equipment areas. Then, based on the chronological order and spatial location of the abnormal events, the fault propagation chain is reconstructed along the fault propagation edges. Subsequently, the consistency of the reconstruction results is verified by combining the correspondence between the fault propagation chain and the fault stage nodes to obtain the evidence consistency evaluation results. Finally, candidate fault propagation chains are selected based on the evidence consistency evaluation results, and the evolution stage of the candidate fault propagation chains is determined, providing a basis for the generation of subsequent inspection tasks and the output of diagnostic results.

[0035] The component-level fault propagation graph includes source component nodes, fault propagation edges, and fault stage nodes. Fault propagation edges are used to represent the direction of fault propagation between different source component nodes. Fault stage nodes are used to represent the evolution stage of the fault propagation chain. Fault propagation edges are associated with time constraints and spatial constraints.

[0036] In one embodiment, a component-level fault propagation diagram is used to describe the propagation relationships that may form after an anomaly occurs inside or in an auxiliary component of the main transformer. The component-level fault propagation diagram includes source component nodes, fault propagation edges, and fault stage nodes. Source component nodes characterize the initial location of the fault, preferably corresponding to key structurally defined and operationally identifiable parts of the main transformer, such as the winding conductor area, winding lead connection area, core area, magnetic shielding area, cooling branch area, bushing area, insulating medium area, and tap changer-related areas. Fault propagation edges describe the direction of anomaly transmission between different source component nodes. For example, after an abnormal heat dissipation occurs in the cooling branch area, it may further cause an excessively high temperature rise in the winding conductor area; after continuous heating in the winding conductor area, it may further cause accelerated aging in the insulating medium area; after a local anomaly occurs in the bushing area, it may extend to the lead connection area.

[0037] The fault propagation edge is not an arbitrary connection, but is pre-established based on the main transformer's structural composition, energy transfer path, dielectric coupling relationship, and historical fault records. During establishment, equipment design drawings, typical fault mechanism data, maintenance cases, and operation and maintenance experience can be jointly analyzed. Fault stage nodes are used to indicate the evolutionary state of the fault propagation chain, preferably divided into initial stage, development stage, and expansion stage. The initial stage indicates that the anomaly has just appeared and its impact range is small; the development stage indicates that the anomaly has persisted and propagated to adjacent parts; the expansion stage indicates that the anomaly has formed a stable impact on multiple related parts.

[0038] Fault propagation edges are also associated with time and spatial constraints. Time constraints specify the allowable time interval between consecutive anomalies, and the principles for setting these constraints can be based on historical health and fault operation data statistics, or on the experience of maintenance personnel in judging the speed of fault evolution. Spatial constraints specify the allowable propagation direction and range of anomalies on the equipment structure. For example, an anomaly in a cooling branch should primarily affect the local temperature rise area corresponding to that branch, rather than directly jumping to an independent part far away from that branch. By introducing time and spatial constraints, the component-level fault propagation diagram can not only represent "which components may propagate to each other," but also "within what time range and along what spatial path," thus providing a clear structured basis for subsequent anomaly event matching and fault propagation chain judgment.

[0039] Based on the component-level fault propagation graph, abnormal events are associated, fault propagation chains are reconstructed, and candidate fault propagation chains are determined according to the consistency of evidence. This includes: matching abnormal events to source component nodes and fault propagation edges in the component-level fault propagation graph; determining the evaluation results that characterize the consistency of evidence based on the correspondence between the number, occurrence order, and spatial location of abnormal events and source component nodes, fault propagation edges, and fault stage nodes; determining candidate fault propagation chains based on the evaluation results, and determining the evolution stages of the candidate fault propagation chains.

[0040] In one embodiment, the association process between abnormal events and component-level fault propagation graphs includes three stages: abnormal event matching, evidence consistency evaluation, and candidate fault propagation chain screening. During abnormal event matching, the abnormal event is first mapped to the source component node in the component-level fault propagation graph based on the preset component domain, spatial location, and abnormality type corresponding to the abnormal event. If the abnormal event mainly manifests as abnormal winding temperature rise, it is preferentially mapped to the winding conductor region; if the abnormal event mainly manifests as abnormal core grounding current and abnormal temperature rise in fixed areas, it is preferentially mapped to the core region; if the abnormal event mainly manifests as abnormal cooler operation and insufficient local heat dissipation, it is preferentially mapped to the cooling branch region.

[0041] After mapping the source component nodes, the order of occurrence of abnormal events is used to determine whether a continuous propagation path can be formed along the fault propagation edge. This considers not only the number of abnormal events but also whether they appear sequentially according to the direction set by the fault propagation edge. If an abnormal event is geographically close but its order of occurrence clearly deviates from the preset propagation direction, it is not considered a continuous event in the same fault propagation chain. Spatial location is also considered; when two abnormal events are adjacent or correspondent in structural location and there is no obvious break in between, they can be determined to have spatial continuity. Based on the number, order of occurrence, and spatial location of abnormal events, and their correspondence with source component nodes, fault propagation edges, and fault stage nodes, an evidence consistency evaluation result is formed.

[0042] The results of the evidence consistency evaluation can be graded, such as high consistency, medium consistency, and low consistency. The grading criteria mainly include three aspects: first, the number of critical nodes covered by the abnormal event; second, whether the abnormal events meet preset time constraints; and third, whether the locations corresponding to the abnormal events meet preset spatial constraints. If multiple abnormal events continuously match the same propagation path, and both the temporal and spatial relationships conform to preset rules, it is judged as high consistency; if only some relationships are met, it is judged as medium consistency; if the matching relationships are scattered or there are obvious conflicts, it is judged as low consistency. After obtaining the evidence consistency evaluation results, the propagation path that meets the preset consistency level requirements is determined as the candidate fault propagation chain, and the evolution stage is determined based on the number of source component nodes already covered by the propagation path, the duration of the abnormality, and the propagation range. When the abnormality is concentrated only in a single source component node, it can be judged as the initial stage; when the abnormality extends outward along adjacent fault propagation edges, it can be judged as the development stage; when the abnormality has formed a continuous impact in multiple locations, it can be judged as the expansion stage. In this way, multiple abnormal events can be organized into a candidate fault propagation chain with sequential and spatial logic without relying on isolated judgments based on a single monitoring quantity, providing direct support for the targeted generation of subsequent inspection tasks.

[0043] Inspection tasks are generated based on the discriminative power of candidate fault propagation chains. New monitoring data is acquired based on the inspection tasks, and candidate fault propagation chains are updated based on the new monitoring data. The target fault propagation chain is determined, and the status assessment results and diagnostic results are output.

[0044] After obtaining candidate fault propagation chains, the evidence of differences between different candidate fault propagation chains is compared to identify key inspection targets that can distinguish between them. Then, inspection tasks are generated based on the key inspection targets, and new monitoring data is acquired according to the inspection tasks. Subsequently, the new monitoring data is used to update the operating condition deviation, component abnormal intensity, and abnormal events, and the evaluation results of the candidate fault propagation chains are reconstructed. When the evaluation results meet the target determination conditions, the target fault propagation chain is determined. Finally, the status assessment results and diagnostic results are output based on the target fault propagation chain, so that inspection, diagnosis, and updating form a closed loop.

[0045] The inspection task is generated based on the distinguishability of the candidate fault propagation chain, including: identifying at least two candidate fault propagation chains with the highest evaluation results; determining the target inspection area and target inspection method based on the difference evidence used to distinguish the at least two candidate fault propagation chains; and generating the inspection task based on the target inspection area and target inspection method.

[0046] In one embodiment, the inspection task is generated based on the need to differentiate between candidate fault propagation chains, rather than performing an indiscriminate review of all equipment areas. This additional constraint ensures that the inspection action directly correlates with evidence of differences between candidate fault propagation chains, avoiding excessively broad inspection scopes or inspection content irrelevant to the current diagnostic ambiguity.

[0047] In practice, candidate fault propagation chains are first sorted from highest to lowest evaluation results, and at least two of the highest-ranking candidate fault propagation chains are selected as the objects to be differentiated. The reason for selecting at least two candidate fault propagation chains is that a single candidate fault propagation chain can only reflect the current optimal judgment and cannot reflect the source of diagnostic ambiguity. Only by analyzing candidate fault propagation chains with similar evaluation results side-by-side can the key differences that need to be further identified through new monitoring data be determined. Then, evidence of differences between the at least two candidate fault propagation chains is extracted.

[0048] Difference evidence can manifest as different locations of anomalies, different propagation sequences, different durations of anomalies, or different auxiliary equipment involved in the anomaly. For example, if one candidate fault propagation chain starts from the cooling branch area and another starts from the winding conductor area, the difference evidence will mainly be reflected in the radiator temperature difference distribution, oil pump operating status, hot spot temperature rise location, and hot spot expansion direction. As another example, if one candidate fault propagation chain starts from the bushing area and another starts from the lead connection area, the difference evidence will mainly be reflected in the bushing root temperature rise, changes in dielectric loss, and localized temperature concentration areas in the lead wires. After identifying the difference evidence, it is mapped to specific equipment areas to form target inspection areas. Target inspection areas can be single locations or combinations of multiple related locations, but each target inspection area should be directly related to the differentiation requirements of the current candidate fault propagation chain.

[0049] Furthermore, the target inspection method is determined based on the type of discrepancy evidence. If the discrepancy evidence mainly manifests as differences in temperature distribution, temperature-related inspection methods are prioritized; if the discrepancy evidence mainly manifests as differences in discharge activity, discharge-related inspection methods are prioritized; if the discrepancy evidence mainly manifests as differences in auxiliary equipment status, equipment status verification inspection methods are prioritized. Finally, the target inspection area, target inspection method, inspection execution period, and data feedback requirements are combined to generate an inspection task, which is then sent to the inspection terminal, monitoring platform, or automatic data acquisition device to obtain targeted new monitoring data during the current diagnostic phase.

[0050] The target inspection method includes at least one of directional infrared rescanning, partial discharge encrypted sampling, cooling system status verification, and accessory operation status verification; new monitoring data is acquired based on the inspection task, and the candidate fault propagation chain is updated based on the new monitoring data, including: acquiring new monitoring data corresponding to the target inspection area; updating the operating condition deviation and component abnormality intensity of the corresponding preset component domain based on the new monitoring data; regenerating abnormal events based on the updated component abnormality intensity, and updating the candidate fault propagation chain based on the regenerated abnormal events.

[0051] In one embodiment, after the inspection task is executed, the newly added monitoring data is used to back-update the current diagnostic results, rather than being saved as isolated independent data. The added limitation of this setting is that the newly added monitoring data is directly incorporated into the existing diagnostic chain, enabling the inspection action to reverse-correct the results of previous anomaly identification and propagation. The target inspection method may include at least one of directional infrared rescanning, partial discharge encrypted sampling, cooling system status verification, and accessory operating status verification.

[0052] Directional infrared rescanning is suitable for scenarios with significant temperature distribution differences. By rescanning the target inspection area at multiple times and angles, it obtains the local temperature rise range, temperature rise center location, and temperature rise duration. Partial discharge encrypted sampling is suitable for scenarios with uncertain discharge activity. By shortening the sampling interval and extending the sampling duration, it obtains the partial discharge pulse density, pulse continuity, and pulse position changes. Cooling system status verification is suitable for scenarios where it is difficult to distinguish between cooling branch abnormalities and heating abnormalities. By verifying the temperature difference between the inlet and outlet of the fan, oil pump, and radiator, as well as the switching status of the cooler, it confirms whether the cooling system is actually involved in the formation of the abnormality. Accessory operation status verification is suitable for scenarios where there are ambiguities in the bushing, tap changer, or lead connection areas. By verifying the accessory operation records, auxiliary monitoring values, and on-site status indicators, it determines whether the abnormality is caused by the accessory components. After acquiring new monitoring data, it is first mapped to the corresponding preset component domain according to the target inspection area, and then the operating condition deviation of the relevant preset component domain is recalculated in combination with the current operating conditions.

[0053] After updating the operating condition deviation, the component anomaly intensity of the corresponding preset component domain is updated. If the newly added monitoring data confirms the continued existence of the original anomaly or the expansion of the anomaly range, the component anomaly intensity is increased; if the newly added monitoring data indicates that the original anomaly is only a short-term disturbance or the location no longer matches, the component anomaly intensity is decreased. After the component anomaly intensity is updated, the anomaly event is regenerated. When regenerating the anomaly event, not only are the new anomalies reflected by the newly added monitoring data retained, but the start time, duration, intensity, and spatial location of the original anomaly event are also synchronously corrected. Finally, based on the regenerated anomaly event, the source component node, fault propagation edge, and fault stage node are rematched, and the evaluation results and evolution stages of the candidate fault propagation chain are updated, so that the diagnostic results can converge in real time with the addition of new monitoring data.

[0054] If the difference between the evaluation results of the candidate fault propagation chain with the highest evaluation result and the candidate fault propagation chain with the second highest evaluation result is less than a preset threshold, the inspection task continues to be generated; if the difference between the evaluation results is greater than or equal to the preset threshold, the candidate fault propagation chain with the highest evaluation result is determined as the target fault propagation chain.

[0055] In one embodiment, the target fault propagation chain is determined based on the difference in evaluation results between candidate fault propagation chains, rather than fixing the diagnostic conclusion immediately after the initial reconstruction. This new constraint uses explicit decision conditions to constrain whether the inspection continues, preventing premature termination or invalid repetition. Specifically, the difference in evaluation results between the candidate fault propagation chain with the highest evaluation result and the candidate fault propagation chain with the second highest evaluation result is compared first.

[0056] The evaluation result difference reflects the degree of distinction between the current optimal judgment and the second-best judgment. A smaller evaluation result difference indicates that at least two candidate fault propagation chains can adequately explain the current anomaly, and the existing evidence is insufficient to form a stable conclusion. A larger evaluation result difference indicates that the current optimal candidate fault propagation chain is significantly better than other candidate fault propagation chains, and the process can proceed to the target fault propagation chain determination stage. A preset threshold is used to distinguish between the two states: "evidence still needs to be supplemented" and "evidence is sufficient." The preset threshold can be set based on the distribution of evaluation result differences for correctly identified samples in historical diagnostic data, or based on the operation and maintenance unit's tolerance for misjudgment risk.

[0057] If the difference in evaluation results is less than a preset threshold, inspection tasks continue to be generated, prioritizing target inspection areas and methods that can further amplify the evidence of discrepancies. At this point, the inspection task does not repeat the content of the previous inspection; instead, it readjusts the inspection focus based on the remaining unresolved diagnostic ambiguities. For example, if the previous inspection mainly covered temperature distribution, but the ambiguity still focuses on the presence of partial discharge activity, the next inspection should shift to more intensive partial discharge sampling. If the difference in evaluation results is greater than or equal to a preset threshold, the current evidence is considered sufficient to support a unique diagnostic conclusion, and the candidate fault propagation chain with the highest evaluation result is identified as the target fault propagation chain. This ensures that the determination of the target fault propagation chain is based on clear evidence convergence conditions, rather than relying on subjective experience, and also provides an executable boundary for the inspection termination condition.

[0058] The status assessment results include at least two of the following: current severity, evolution rate, and diagnostic uncertainty. The diagnostic results include the source component, fault propagation path, and evolution stage corresponding to the target fault propagation chain.

[0059] In one embodiment, the status assessment results and diagnostic results are set for the operational status description and fault mechanism description, respectively. The added constraint of this setting is to distinguish the functional boundaries between the status assessment results and the diagnostic results, enabling subsequent maintenance personnel to simultaneously obtain risk information and location information. The status assessment results may include at least two of the following: current severity, evolution rate, and diagnostic uncertainty.

[0060] The current severity level reflects the impact of the anomaly corresponding to the target fault propagation chain on the operational safety of the main transformer at the current moment. It can be classified as general, relatively severe, and serious, or as low-risk, medium-risk, and high-risk. The criteria for setting the current severity level can include the intensity of the component anomaly, the duration of the anomaly event, the coverage area of ​​the anomaly, and whether the anomaly affects critical components. Evolution speed reflects how quickly the anomaly progresses from the initial stage to the development or expansion stage. It can be determined comprehensively based on the magnitude of changes in component anomaly intensity, the number of anomaly events, and the extension speed of the propagation chain within a continuous time window. Diagnostic uncertainty reflects whether there is still a close competition relationship between the target fault propagation chain and other candidate fault propagation chains. The lower the diagnostic uncertainty, the more stable the current conclusion; the higher the diagnostic uncertainty, the more attention needs to be paid to subsequent new monitoring data.

[0061] The diagnostic results include the source component, the fault propagation path, and the evolution stage corresponding to the target fault propagation chain. The source component indicates the most likely starting point of the anomaly, the fault propagation path explains how the anomaly spreads from the starting point to other related parts, and the evolution stage indicates whether the current fault is in its initial, developing, or expanding state. During output, both the status assessment results and the diagnostic results can be sent to a monitoring platform, alarm terminal, or operation and maintenance decision-making system. The status assessment results are more suitable for determining alarm levels and handling priorities, while the diagnostic results are more suitable for guiding subsequent inspections, maintenance, and fault verification. This approach creates a complete closed loop between inspection, diagnosis, and result output, ensuring that the output reflects both the current risk level and the specific path of fault formation and propagation.

[0062] In one specific embodiment, a 220 kV three-phase oil-immersed main transformer with a rated capacity of 180 MVA was selected as the implementation object. The cooling method was a combined cooling system of fans and oil pumps. The sampling period was from 12:00 to 14:30 on a certain day, with a sampling cycle of 10 minutes. The online monitoring data included load current, top oil temperature, estimated winding hot spot temperature, cooler switching status, oil pump operating current, core grounding current, bushing dielectric loss, infrared thermography results, partial discharge count, and dissolved gas in the oil. The current operating conditions were medium-high load, dual fans in operation, single oil pump in operation, stable tap position, and ambient temperature of 29.1°C to 29.8°C. A benchmark operating condition was established based on historical data from the past 90 days under similar conditions, showing no alarms, maintenance, or fault records.

[0063] Under this operating condition, from 12:00 to 14:30, the top oil temperature reference rose from 61.2 degrees Celsius to 69.1 degrees Celsius, and the measured value rose from 61.3 degrees Celsius to 73.1 degrees Celsius; the hot spot temperature reference rose from 74.0 degrees Celsius to 81.5 degrees Celsius, and the measured value rose from 74.1 degrees Celsius to 90.8 degrees Celsius. By 14:10, the top oil temperature was 3.7 degrees Celsius higher than the operating condition reference, and the estimated hot spot temperature was 8.3 degrees Celsius higher than the operating condition reference. Figure 2 This is a comparison chart of the baseline operating condition and the measured temperature. The chart is drawn by combining the estimated top oil temperature and hot spot temperature every 10 minutes during the sampling period of this embodiment with the baseline operating condition formed from health data under the same conditions. Figure 2 As shown, after 12:50, both measured temperature curves began to stabilize above the baseline curve, and the rise rate of the hot spot temperature measured curve was significantly faster than that of the top oil temperature measured curve, indicating that the anomaly first manifested as local heat accumulation, and then gradually spread to the overall oil temperature.

[0064] During the deviation calculation phase, the monitored characteristics are divided into the cooling domain, winding domain, insulation medium domain, bushing and lead domain, and magnetic circuit domain. For ease of unified judgment, the abnormal strength of the component is calculated using the following formula:

[0065] in, For abnormal strength of components, For the first Normalized operating condition deviation of each monitoring characteristic, For the first The weights of each monitoring feature are assigned, and the sum of the weights is 1.

[0066] Taking 14:10 as an example, three key characteristics of the cooling zone are selected for calculation. First, the measured surface temperature of radiator No. 2 is 76.9 degrees Celsius, the operating condition baseline is 64.3 degrees Celsius, the allowable fluctuation range is 3.0 degrees Celsius, and the normalized operating condition deviation is 4.20. Second, the measured oil pump operating current is 5.1 amps, the operating condition baseline is 7.4 amps, the allowable fluctuation range is 0.7 amps, and the normalized operating condition deviation is 3.29. Third, the top layer oil temperature cooling response deviation is 3.7 degrees Celsius, the allowable fluctuation range is 1.8 degrees Celsius, and the normalized operating condition deviation is 2.06. With corresponding weights of 0.45, 0.35, and 0.20, the abnormal intensity of the cooling zone components is 3.45. The winding domain was calculated using hotspot temperature deviation, ethylene growth deviation, and local temperature rise expansion deviation. At 14:10, the normalized operating condition deviation of the hotspot temperature was 3.32. The ethylene concentration increased from 9.1 μL / L at 12:00 to 13.8 μL / L at 14:10, corresponding to a normalized operating condition deviation of 2.12. The local temperature rise expansion deviation was 1.76. With weights of 0.50, 0.30, and 0.20 respectively, the abnormal strength of the winding domain components was 2.67. The abnormal strength of the insulation dielectric domain components was 1.41, the bushing and lead domain was 0.24, and the magnetic circuit domain was 0.18. The abnormal judgment threshold was set at 1.80, based on a safety margin of 10% above the 95th percentile of the healthy sample under the same operating condition (1.62). It is evident that both the cooling domain and the winding domain exceeded the threshold, with the cooling domain exceeding it earlier. Figure 3 This is a timing diagram of component anomaly events. The diagram is drawn by plotting the start and end times, duration, and peak intensity of component anomalies exceeding the anomaly detection threshold in each preset component domain. For example... Figure 3 As shown, the cooling domain anomaly started at 12:40 and lasted until 14:30, the winding domain anomaly started at 12:50, and the insulation medium domain anomaly started at 13:20. However, the bushing and lead domain and the magnetic circuit domain only showed short-term weak anomalies and did not have dominant propagation characteristics.

[0067] During the fault propagation chain reconstruction phase, three main candidate fault propagation chains were established. The first was the cooling degradation chain, namely, abnormal cooling branch, insufficient heat exchange, excessively high local temperature rise in the winding, and heating of the insulating medium. The second was the local heating chain in the winding, namely, the formation of local heating points in the winding, the expansion of hot spots, and heating of the insulating medium. The third was the bushing abnormality chain, namely, local abnormality in the bushing, heating of the leads, and expansion of local temperature rise. After evaluating the consistency of evidence based on the number, sequence, and spatial location of abnormal events, the evaluation results of the three candidate fault propagation chains were 0.83, 0.76, and 0.32, respectively. The difference between the evaluation results of the first and second chains was only 0.07, which was less than the preset threshold of 0.10, therefore, the targeted inspection phase was initiated. The target inspection areas were determined to be the left radiator group, the high-voltage side bushing area, and the tank wall area corresponding to the winding hot spots; the target inspection methods were determined to be directional infrared rescanning, partial discharge encrypted sampling, and cooling system status verification. Figure 4 This is a schematic diagram of the infrared temperature rise distribution in the target inspection area. The diagram was obtained from a directional infrared rescan performed at 14:18, and was drawn after being divided and labeled according to the temperature measurement areas on the equipment surface. Figure 4 As shown, the surface temperature of radiator No. 2 in the left-side radiator group reached 76.9 degrees Celsius, significantly higher than the 63.8 degrees Celsius and 64.5 degrees Celsius of radiators No. 1 and No. 3, respectively. The surface temperature of the main transformer body was 72.6 degrees Celsius, the oil temperature on the top of the tank was 73.1 degrees Celsius, and the three measuring points in the high-voltage side bushing area were 68.4 degrees Celsius, 69.1 degrees Celsius, and 68.8 degrees Celsius, respectively, without forming an independent high-temperature concentration area. This result indicates that the high-temperature concentration area is highly consistent with the spatial location of the cooling branch, but inconsistent with the bushing abnormal chain.

[0068] New monitoring data was continuously updated in the diagnostic process. Partial discharge intensive sampling results showed that the pulse count fluctuated from 26 pulses per minute to 31 pulses per minute between 12:00 and 14:30, reaching 28 pulses per minute at 14:10, close to the upper limit of 30 pulses per minute for healthy samples but not significantly exceeded; the acetylene concentration increased from 0.4 μL / L to 0.5 μL / L, with no significant change. Cooling system status verification results showed that the inlet and outlet temperature difference of the oil flow branch corresponding to radiator No. 2 was only 4.1 degrees Celsius, while the inlet and outlet temperature differences of the branches corresponding to adjacent radiators No. 1 and No. 3 were 7.6 degrees Celsius and 7.8 degrees Celsius, respectively, indicating a decrease in local heat dissipation capacity. After incorporating the new monitoring data into the update, the cooling degradation chain evaluation result increased from 0.83 to 0.91, the winding local heating chain decreased from 0.76 to 0.58, and the bushing abnormality chain decreased from 0.32 to 0.21. Figure 5 This is a comparison chart of the evaluation results of candidate fault propagation chains before and after the inspection. The chart is drawn from the evaluation results of three candidate fault propagation chains before and after the inspection. Figure 5As shown, the cooling degradation chain after inspection is significantly higher than the other two candidate fault propagation chains. The difference between the highest evaluation result and the second highest evaluation result reaches 0.33, which exceeds the preset threshold of 0.10. Therefore, the cooling degradation chain is identified as the target fault propagation chain, and the evolution stage is determined to be the development stage.

[0069] Based on the output status assessment and diagnostic results of the target fault propagation chain, the current severity is set at 0.82, indicating a relatively severe condition; the evolution rate is set at 0.27 per hour, indicating that the anomaly is still developing; and the diagnostic uncertainty is set at 0.12, indicating that the newly added monitoring data has reduced the main diagnostic ambiguity to a low level. The diagnostic results show that the source component is the cooling branch corresponding to radiator #2 on the left side. The fault propagation path is: abnormal cooling branch, insufficient heat exchange, excessive winding temperature rise, and heating of the insulation medium. The evolution stage is the development stage. Based on these results, on-site repairs were carried out on the radiator #2 branch and oil circuit verification measures were implemented. After 24 hours of retesting following the repairs, the top oil temperature recovered to within 0.8 degrees Celsius of the baseline under the same operating conditions, and the hot spot temperature recovered to within 1.6 degrees Celsius of the baseline under the same operating conditions. This demonstrates that the method can locate the fault before local overheating develops into a global anomaly and converge the candidate fault propagation chain to a single target fault propagation chain through a closed-loop inspection process, demonstrating good practical application value.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0071] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for intelligent inspection and condition assessment diagnosis of a main transformer, characterized in that, The method includes: The operating condition baseline is determined based on the multi-source online monitoring data and operating conditions of the main transformer. The operating condition deviation of the multi-source online monitoring data of the main transformer is calculated according to the preset component domain, the abnormal intensity of the component is determined based on the operating condition deviation, and an abnormal event is generated according to the change of the abnormal intensity of the component. Based on the component-level fault propagation graph, the abnormal events are associated, the fault propagation chain is reconstructed, and candidate fault propagation chains are determined based on the consistency of evidence. An inspection task is generated based on the distinguishability of the candidate fault propagation chain. New monitoring data is obtained based on the inspection task, and the candidate fault propagation chain is updated based on the new monitoring data. The target fault propagation chain is determined, and the status assessment result and diagnosis result are output.

2. The method according to claim 1, characterized in that, The determination of the operating condition benchmark based on the multi-source online monitoring data of the main transformer and the operating conditions includes: The multi-source online monitoring data of the main transformer are subjected to time synchronization processing and data reliability assessment. The operating conditions are divided according to the load level, cooling status, tap position and ambient temperature. The operating condition baselines corresponding to each operating condition are determined based on historical health operation data.

3. The method according to claim 2, characterized in that, The step of calculating the operating condition deviation of the multi-source online monitoring data of the main transformer according to the preset component domain includes: Extract the monitoring features corresponding to each preset component domain; Based on the aforementioned operating condition benchmark, determine the reference values ​​of each monitoring feature under the current operating conditions; The operating condition deviation corresponding to each preset component domain is calculated based on the observed values ​​of each monitoring feature and the reference values, and the operating condition deviation is normalized. The preset component domains include winding domain, magnetic circuit domain, cooling domain, bushing and lead domain, and insulating medium domain.

4. The method according to claim 3, characterized in that, The process of determining the abnormal strength of a component based on the operating condition deviation and generating an abnormal event based on the change in the abnormal strength of the component includes: The initial component anomaly intensity is determined based on the operating condition deviation and the data reliability assessment results corresponding to each monitoring feature in each preset component domain; Based on the temporal and spatial proximity relationships of each monitored feature within the same preset component domain, the initial component anomaly intensity is correlated and corrected to obtain the component anomaly intensity. An abnormal event is generated when the abnormal intensity of the component continuously exceeds a preset threshold and the trend of change changes. The abnormal event is used to characterize the start time, duration, intensity, and spatial location of the corresponding preset component domain.

5. The method according to claim 1, characterized in that, The component-level fault propagation graph includes source component nodes, fault propagation edges, and fault stage nodes; The fault propagation edge is used to represent the fault propagation direction between different source component nodes; The fault stage node is used to represent the evolutionary stage of the fault propagation chain; The fault propagation edge association has both time and space constraints.

6. The method according to claim 5, characterized in that, The process of associating the abnormal events with the component-level fault propagation graph, reconstructing the fault propagation chain, and determining candidate fault propagation chains based on evidence consistency includes: The abnormal event is matched to the source component node and fault propagation edge in the component-level fault propagation graph; Based on the correspondence between the number, order of occurrence, and spatial location of the abnormal events and the source component nodes, the fault propagation edges, and the fault stage nodes, the evaluation result representing the consistency of the evidence is determined. The candidate fault propagation chain is determined based on the evaluation results, and the evolution stage of the candidate fault propagation chain is determined.

7. The method according to claim 6, characterized in that, The step of generating inspection tasks based on the discriminative power of the candidate fault propagation chain includes: Identify at least two candidate fault propagation chains with the highest evaluation results; The target inspection area and target inspection method are determined based on the differential evidence used to distinguish the at least two candidate fault propagation chains. The inspection task is generated based on the target inspection area and the target inspection method.

8. The method according to claim 7, characterized in that, The target inspection method includes at least one of directional infrared rescanning, partial discharge encrypted sampling, cooling system status verification, and accessory operation status verification. The step of acquiring the new monitoring data based on the inspection task and updating the candidate fault propagation chain according to the new monitoring data includes: Acquire the newly added monitoring data corresponding to the target inspection area; The operating condition deviation and component abnormality intensity of the corresponding preset component domain are updated based on the newly added monitoring data; The abnormal events are regenerated based on the updated component abnormality intensity, and the candidate fault propagation chain is updated based on the regenerated abnormal events.

9. The method according to claim 8, characterized in that, If the difference between the evaluation results of the candidate fault propagation chain with the highest evaluation result and the candidate fault propagation chain with the second highest evaluation result is less than a preset threshold, the inspection task continues to be generated. If the difference in the evaluation results is greater than or equal to the preset threshold, the candidate fault propagation chain with the highest evaluation result is determined as the target fault propagation chain.

10. The method according to claim 9, characterized in that, The state assessment results include at least two of the following: current severity, evolution rate, and diagnostic uncertainty. The diagnostic results include the source component, fault propagation path, and evolution stage corresponding to the target fault propagation chain.

Images