A multi-source heterogeneous data fusion analysis method for intelligent operation and maintenance of a power distribution station

Abstract

This invention relates to the field of intelligent operation and maintenance technology for power distribution substations, and discloses a method for multi-source heterogeneous data fusion analysis in intelligent operation and maintenance of power distribution substations. This method acquires multi-source heterogeneous data from multiple data sources within the power distribution substation and converts it into unified event data; calculates the dynamic credibility of the data sources based on data integrity scores, timeliness scores, fluctuation stability scores, and historical consistency scores; corrects the acquisition time based on historical transmission delays; generates an adaptive fusion time window based on the type of potential hazard, and aggregates candidate potential hazard events within this time window; calculates an initial fusion risk value by combining anomaly probability and data source dynamic credibility, and then obtains a corrected fusion risk value through a conflict penalty coefficient, outputting the hazard level and hazard evidence chain, and updating it based on the handling feedback results. This invention improves the accuracy and interpretability of multi-source data fusion analysis in power distribution substations.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power distribution substations, specifically a method for multi-source heterogeneous data fusion and analysis in intelligent operation and maintenance of power distribution substations. Background Technology

[0002] With the development of intelligent construction in power distribution substations, these substations are typically equipped with various data sources, including electrical quantity acquisition devices, equipment temperature measuring devices, partial discharge sensors, temperature and humidity sensors, water immersion sensors, smoke sensors, video cameras, access control devices, and work order management systems. Existing intelligent operation and maintenance systems for power distribution substations generally access these multi-source data sources through edge gateways or monitoring platforms. Based on threshold judgment, rule matching, video recognition, or multi-source data fusion, they monitor and issue alarms for the equipment status, environmental status, security status, and personnel operation status within the substation, thereby achieving centralized monitoring and assisted operation and maintenance of the power distribution substation.

[0003] However, existing technologies, when performing multi-source data fusion analysis, typically aggregate data from different data sources directly based on data reception time or a fixed time window. This fails to fully consider issues such as different sampling frequencies, transmission delays, data gaps, sensor drift, video obstruction, and differences in the reliability of historical alarms among different data sources. As a result, multi-source evidence of the same hidden danger may be split into multiple alarms, data from different hidden dangers may be mistakenly aggregated into the same event, or low-reliability data may have an excessive impact on the fusion results, leading to false alarms, missed alarms, and duplicate alarms, thus reducing the accuracy and interpretability of the intelligent operation and maintenance results of power distribution substations. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations. This method solves the problem of low accuracy in fusion analysis caused by data source transmission delays, inconsistent sampling frequencies, and varying degrees of data source reliability in existing intelligent operation and maintenance systems for power distribution substations.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations, comprising: Acquire multi-source heterogeneous data generated by multiple data sources within the power distribution station, and convert the multi-source heterogeneous data into unified event data; Based on the data integrity score, timeliness score, fluctuation stability score, and historical consistency score corresponding to the unified event data, the dynamic reliability of the data source to which the unified event data belongs is calculated. Based on the reception time of the unified event data and the historical transmission delay of the data source to which the unified event data belongs, the collection time of the unified event data is corrected to obtain the corrected collection time of the unified event data. Based on the type of hazard, the historical transmission delay, and the preset evolution cycle parameter, an adaptive fusion time window corresponding to the type of hazard is generated; Based on the corrected acquisition time, the unified event data is aggregated according to the device object, spatial region and the type of hazard within the adaptive fusion time window to generate candidate hazard events; Based on the anomaly probability of each unified event data in the candidate potential event and the dynamic reliability of the data source, the initial fusion risk value of the candidate potential event is calculated; Based on the degree of conflict between the unified event data in the candidate hidden danger events, a conflict penalty coefficient is generated, and the initial fusion risk value is corrected according to the conflict penalty coefficient to obtain a corrected fusion risk value; The hazard level and hazard evidence chain are output based on the corrected fusion risk value, and the historical consistency score and the adaptive fusion time window are updated based on the handling feedback results.

[0006] Preferably, the multi-source heterogeneous data includes electrical operation data, equipment status data, environmental monitoring data, video image data, and access control work order data; the unified event data includes data source number, equipment object, spatial area, indicator type, indicator value, collection time, reception time, initial confidence level, and hazard type.

[0007] Preferably, the dynamic credibility of the data source is obtained by weighting the data integrity score, the timeliness score, the volatility stability score, and the historical consistency score, wherein the data integrity score, the timeliness score, the volatility stability score, and the historical consistency score are each configured with a credibility weight coefficient.

[0008] Preferably, the data integrity score is determined based on the missing proportion of the unified event data within a preset statistical period; the timeliness score is determined based on the historical transmission delay; the fluctuation stability score is determined based on the number of abnormal jumps in the indicator value within a preset statistical period; and the historical consistency score is determined based on the number of times the candidate hidden danger events in which the data source participates are consistent with the handling feedback results.

[0009] Preferably, the step of obtaining the corrected acquisition time includes: acquiring multiple transmission delay samples of the same data source within a historical statistical period; calculating the historical transmission delay of the data source based on the multiple transmission delay samples; and subtracting the historical transmission delay from the receiving time to obtain the corrected acquisition time.

[0010] Preferably, the adaptive fusion time window is determined based on the basic time window corresponding to the hazard type, the maximum historical transmission delay among the multiple data sources associated with the hazard type, and the preset evolution cycle parameter, and the maximum historical transmission delay and the preset evolution cycle parameter are weighted and corrected by a time window adjustment coefficient.

[0011] Preferably, the step of generating the candidate potential hazard event includes: grouping multiple unified event data that are the same in terms of equipment object, spatial region, and hazard type, and whose correction collection time falls within the same adaptive fusion time window into the same event set; when the number of unified event data in the event set reaches a preset data quantity threshold, or the data source type corresponding to the unified event data in the event set reaches a preset data source type threshold, the event set is determined as the candidate potential hazard event.

[0012] Preferably, the initial fusion risk value is obtained by weighting the anomaly probability corresponding to each of the unified event data in the candidate hidden danger events and the dynamic credibility of the data source, wherein the dynamic credibility of the data source is used to determine the contribution weight of the corresponding anomaly probability in the fusion calculation.

[0013] Preferably, the corrected fusion risk value is obtained by attenuating and correcting the initial fusion risk value using the conflict penalty coefficient, wherein the larger the conflict penalty coefficient, the greater the attenuation of the corrected fusion risk value relative to the initial fusion risk value.

[0014] Preferably, the chain of evidence for potential hazards includes the hazard object, hazard type, triggering data source, triggering time, main evidence data, auxiliary evidence data, low-credibility evidence data, and handling recommendations; The feedback results include feedback results on actual potential hazards, feedback results on false alarms, and feedback results on missed alarms. When the handling feedback result is the actual hidden danger feedback result, the historical consistency score corresponding to the data source participating in the candidate hidden danger event is increased; When the handling feedback result is the false alarm feedback result, the historical consistency score corresponding to the data source participating in the candidate hidden danger event is reduced; When the handling feedback result is the missed report feedback result, adjust the basic time window or the preset evolution cycle parameter corresponding to the hidden danger type, and regenerate the adaptive fusion time window.

[0015] This invention provides a multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations. It has the following beneficial effects: 1. This invention calculates the dynamic reliability of data sources through data integrity scoring, timeliness scoring, fluctuation stability scoring, and historical consistency scoring. It also combines historical transmission delay to obtain a corrected collection time and generates an adaptive fusion time window based on the type of hidden danger. This allows data sources with different sampling frequencies, transmission delays, and reliability levels to be fused and analyzed within a more reasonable time range. This reduces the situation where the same hidden danger is split into multiple alarms or different hidden dangers are mistakenly aggregated into the same event, thereby improving the accuracy of intelligent operation and maintenance fusion analysis of power distribution substations.

[0016] 2. When generating candidate potential incidents, this invention aggregates unified incident data according to equipment object, spatial area, and incident type, enabling multi-source heterogeneous data to form an incident set around a specific incident object. This is beneficial for merging environmental anomalies, equipment anomalies, video recognition results, and access control work order information into the same candidate potential incident, reducing the probability of repeated alarms.

[0017] 3. This invention calculates the initial fusion risk value based on the anomaly probability of each unified event data in the candidate hidden danger events and the dynamic credibility of the data source, so that the high credibility data source has a higher contribution to the fusion result, and the impact of the low credibility data source on the fusion result is weakened, which can reduce the interference of sensor drift, video occlusion, data missing or communication anomaly on the hidden danger judgment result.

[0018] 4. This invention corrects the initial fusion risk value by using a conflict penalty coefficient. When there is a significant conflict between the unified event data in the candidate hidden event, the final output corrected fusion risk value is reduced, thereby reducing false alarms caused by isolated abnormal data or single abnormal data.

[0019] 5. This invention outputs the hazard level and the hazard evidence chain. The hazard evidence chain includes the hazard object, hazard type, triggering data source, triggering time, main evidence data, auxiliary evidence data, low-reliability evidence data, and handling suggestions, enabling operation and maintenance personnel to clearly identify the alarm source and judgment basis, and improve the interpretability and handling efficiency of the intelligent operation and maintenance results of the power distribution station.

[0020] 6. This invention updates the historical consistency score and adaptive fusion time window based on the feedback results of the handling, enabling the system to continuously correct the reliability of the data source and the fusion time range corresponding to the type of hidden danger based on the actual hidden danger feedback results, false alarm feedback results, and missed alarm feedback results, thereby adapting to the communication conditions, equipment status, and environmental characteristics of different power distribution stations and improving long-term operational stability. Attached Figure Description

[0021] Figure 1 This is a flowchart of a multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution stations according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention provides a multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations. This method is used in intelligent operation and maintenance scenarios of power distribution substations to uniformly process electrical operation data, equipment status data, environmental monitoring data, video image data, and access control work order data. Before fusion analysis, dynamic reliability assessment and time consistency correction are performed on each data source, thereby reducing false alarms, missed alarms, and duplicate alarms caused by data source transmission delays, inconsistent sampling frequencies, sensor drift, video obstruction, or data loss.

[0024] In this embodiment, the substation includes a transformer room, switchgear room, cable tray, low-voltage distribution area, communication equipment area, and entrance / exit area. Data sources can include voltage acquisition terminals, current acquisition terminals, switchgear temperature measuring devices, cable joint temperature measuring devices, partial discharge sensors, temperature and humidity sensors, water immersion sensors, smoke sensors, video cameras, access control devices, work permit systems, and inspection record systems. Each data source connects to the substation's intelligent operation and maintenance platform via a station edge gateway or communication network.

[0025] Reference Figure 1 The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations in this embodiment includes the following steps.

[0026] Acquire multi-source heterogeneous data generated by multiple data sources within the power distribution station and convert the multi-source heterogeneous data into unified event data.

[0027] Specifically, during the operation of the power distribution station, multi-source heterogeneous data is acquired from multiple data sources. This multi-source heterogeneous data includes electrical operation data, equipment status data, environmental monitoring data, video image data, and access control work order data.

[0028] The electrical operation data includes at least one of voltage, current, load factor, power factor, and three-phase imbalance rate; the equipment status data includes at least one of switchgear temperature, cable joint temperature, transformer temperature, partial discharge intensity, and equipment noise intensity; the environmental monitoring data includes at least one of ambient temperature, ambient humidity, water immersion status, smoke concentration, oxygen concentration, and harmful gas concentration; the video image data includes at least one of cabinet exterior images, ground water accumulation images, smoke images, and personnel behavior images; and the access control work order data includes at least one of personnel entry and exit records, work order information, operation ticket information, and patrol records.

[0029] Because different data sources have different data formats, this embodiment first converts multi-source heterogeneous data into unified event data. Unified event data includes data source number, device object, spatial region, indicator type, indicator value, collection time, reception time, initial confidence level, and hazard type.

[0030] For example, the data uploaded by the temperature and humidity sensor can be converted into the following unified event data: the data source number is the temperature and humidity sensor number, the device object is the corresponding switch cabinet, the space area is the switch cabinet room, the index type is the ambient humidity, the index value is the humidity value, the acquisition time is the sensor acquisition time, the reception time is the platform reception time, the initial confidence level is the confidence level corresponding to the sensor's factory calibration or historical operating status, and the hazard type is the condensation hazard type.

[0031] For example, when a video camera detects an image of water accumulation on the ground, the video recognition result can be converted into unified event data: the data source number is the camera number, the device object is the device object associated with the camera's field of view, the spatial region is the region corresponding to the camera's field of view, the index type is the image recognition result, the index value is the probability of water accumulation on the ground or the state of water accumulation on the ground, the acquisition time is the video frame time, the reception time is the platform reception time, the initial confidence is the confidence output by the video recognition model, and the hazard type is the water immersion hazard type.

[0032] This step enables the conversion of data from different protocols, formats, and sampling methods into event-based data that can be processed uniformly, providing a data foundation for subsequent credibility calculations and time window fusion.

[0033] After obtaining the unified event data, the dynamic reliability of the data source to which the unified event data belongs is calculated based on the data integrity score, timeliness score, fluctuation stability score, and historical consistency score corresponding to the unified event data.

[0034] Specifically, the data integrity score is used to characterize the data missingness of a corresponding data source within a preset statistical period. The preset statistical period can be set to 5 minutes, 10 minutes, 30 minutes, or 1 hour. If a data source is required to upload 100 data entries within the preset statistical period, but actually uploads 95, the data integrity score can be calculated based on the missing data ratio.

[0035] The timeliness score is used to characterize the timeliness of data transmission from the corresponding data source. If the delay between data collection and platform reception from a data source is small, the timeliness score is high; if a data source has significant communication delays or cached uploads, the timeliness score is low.

[0036] The fluctuation stability score is used to characterize whether there are abnormal jumps in the indicator values ​​of the corresponding data source within a preset statistical period. If a temperature and humidity sensor experiences sudden increases or decreases that do not conform to the laws of physical change within a short period of time, the fluctuation stability score of that data source is considered to have decreased. If a video camera's recognition results change frequently due to obstruction, backlighting, or insufficient lighting at night, the fluctuation stability score of the corresponding data source can also be reduced.

[0037] Historical consistency score is used to characterize the degree of consistency between the response results and the handling feedback after a corresponding data source participates in candidate potential incidents. If multiple candidate potential incidents in which a data source participates are confirmed by operations and maintenance personnel to be genuine potential incidents, the historical consistency score of that data source will be increased; if multiple candidate potential incidents in which a data source participates are confirmed to be false alarms, the historical consistency score of that data source will be decreased.

[0038] In one alternative implementation, the dynamic reliability of the data source can be calculated as follows: C = a·I + b·T + c·S + d·H; Where C represents the dynamic credibility of the data source, I represents the data integrity score, T represents the timeliness score, S represents the fluctuation stability score, H represents the historical consistency score, and a, b, c, and d represent the credibility weight coefficients, respectively.

[0039] The credibility weighting coefficients can be pre-configured according to different data source types. For example, for video cameras, the weights corresponding to fluctuation stability scores and historical consistency scores can be increased; for temperature and humidity sensors, the weights corresponding to data integrity scores and fluctuation stability scores can be increased; and for access control devices, the weights corresponding to timeliness scores and historical consistency scores can be increased.

[0040] By calculating the dynamic credibility of the data source, this embodiment can avoid treating all data sources as equally credible, thereby reducing the impact of a single faulty sensor or low-quality video recognition result on the fusion analysis result.

[0041] Based on the reception time of the unified event data and the historical transmission delay of the data source to which the unified event data belongs, the collection time of the unified event data is corrected to obtain the corrected collection time of the unified event data. Specifically, different data sources have different data upload paths, resulting in varying degrees of transmission delay. For example, electrical operation data can typically be uploaded within seconds, while video image data may require edge detection before uploading, and access control work order data may be synchronized with a delay in the form of events. If the reception time is directly used for fusion, it is easy to misclassify multi-source evidence of the same potential hazard process into different time periods.

[0042] Therefore, this embodiment corrects the collection time of unified event data based on the reception time of unified event data and the historical transmission delay of the data source to which the unified event data belongs, thus obtaining the corrected collection time of unified event data.

[0043] Specifically, multiple transmission delay samples from the same data source within a historical statistical period are obtained. Each transmission delay sample can be determined by the difference between the platform's reception time and the data source's acquisition time. Then, the historical transmission delay of the data source is calculated based on multiple transmission delay samples.

[0044] In one alternative implementation, the average of multiple transmission delay samples can be used as the historical transmission delay; in another alternative implementation, after removing abnormal transmission delay samples, the median of the remaining transmission delay samples can be used as the historical transmission delay. Using the median can reduce the impact of occasional network congestion on the historical transmission delay calculation results.

[0045] After obtaining the historical transmission delay, subtract the historical transmission delay from the receiving time to obtain the corrected acquisition time. That is: Tc = Tr - Δt; Where Tc represents the corrected acquisition time, Tr represents the reception time, and Δt represents the historical transmission delay.

[0046] This step enables unified event data from different data sources to more closely approximate the actual occurrence time in the time dimension, thereby improving the accuracy of subsequent adaptive fusion of events within the time window.

[0047] Based on the hazard type, historical transmission delay, and preset evolution cycle parameters, an adaptive fusion time window corresponding to the hazard type is generated; Specifically, different types of hazards have different evolution rates. Hazards such as personnel crossing boundaries and unauthorized opening of doors are usually short-term events; hazards such as smoke and water immersion have medium evolution cycles; and hazards such as condensation, temperature rise, and partial discharge usually have a longer, gradual evolution process. Therefore, this embodiment generates an adaptive fusion time window corresponding to the hazard type based on the hazard type, historical transmission delay, and preset evolution cycle parameters.

[0048] The types of hazards may include at least one of the following: condensation hazard, water immersion hazard, smoke hazard, temperature rise hazard, partial discharge hazard, personnel crossing boundaries hazard, access control malfunction hazard, and load malfunction hazard.

[0049] The preset evolution period parameter is used to characterize the time characteristics of a certain type of hazard from initial signs to obvious anomalies. For example, the preset evolution period parameter for condensation hazard type can be greater than that for personnel crossing boundary hazard type; the preset evolution period parameter for temperature rise hazard type can be set according to the load change pattern; and the preset evolution period parameter for water immersion hazard type can be set according to the response time of water immersion sensor and video recognition response time.

[0050] In one alternative implementation, the adaptive fusion time window can be generated as follows: W = W0 + k1·D + k2·P; Where W represents the adaptive fusion time window, W0 represents the basic time window corresponding to the hazard type, D represents the maximum historical transmission delay among multiple data sources associated with the hazard type, P represents the preset evolution cycle parameter, and k1 and k2 represent the time window adjustment coefficients, respectively.

[0051] For example, for the risk of personnel crossing boundaries, the basic time window can be set to 10 to 30 seconds; for the risk of water immersion, the basic time window can be set to 30 to 120 seconds; and for the risk of condensation, the basic time window can be set to 5 to 30 minutes. If the corresponding data source has a large historical transmission delay, the adaptive fusion time window is extended by D; if the corresponding risk type has a long evolution cycle, the adaptive fusion time window is further extended by P.

[0052] By generating an adaptive fusion time window, this embodiment can avoid the problems of incorrect or missed aggregation caused by using a fixed fusion time window for all types of potential problems.

[0053] After obtaining the corrected acquisition time and the adaptive fusion time window, the unified event data is aggregated according to the device object, spatial region and hazard type within the adaptive fusion time window based on the corrected acquisition time to generate candidate hazard events.

[0054] Specifically, multiple unified event data that are the same in terms of equipment, spatial area, and hazard type, and whose correction and collection times fall within the same adaptive fusion time window, are grouped into the same event set.

[0055] When the number of unified event data in the event set reaches a preset data quantity threshold, or when the data source type corresponding to the unified event data in the event set reaches a preset data source type threshold, the event set is identified as a candidate potential event.

[0056] For example, regarding a condensation hazard type for a certain switchgear, if within the same adaptive fusion time window, there are instances of increased ambient humidity, abnormal cabinet temperature differences, a slight increase in partial discharge intensity, and suspected condensation reflections in video images, then the aforementioned unified event data are grouped into the same event set. When this event set contains at least three data source types, the event set is identified as a candidate hazard event corresponding to that switchgear.

[0057] For example, regarding a type of water immersion hazard in a cable trench, if a water immersion sensor is triggered within the same adaptive fusion time window, and video images simultaneously detect water accumulation on the ground, and access control records for the same area show that personnel have recently entered, then the aforementioned unified event data can be aggregated into a single candidate hazard event.

[0058] This step allows for the merging of relevant evidence from different data sources for the same potential hazard, preventing the same hazard from being split into multiple independent alarms.

[0059] After generating candidate potential events, the initial fusion risk value of the candidate potential events is calculated based on the anomaly probability of each unified event data in the candidate potential events and the dynamic credibility of the data source.

[0060] Specifically, the anomaly probability represents the degree to which a set of uniform event data supports anomalies of a corresponding hazard type. For numerical indicators, the anomaly probability can be determined based on the deviation between the indicator value and a preset threshold; for video image data, the anomaly probability can be determined based on the recognition confidence output by the video recognition model; and for access control work order data, the anomaly probability can be determined based on the matching relationship between access control events and work tickets / operation tickets.

[0061] In one alternative implementation, the initial fusion risk value can be calculated as follows: R0 = Σ(Ci·Pi) / ΣCi; Where R0 represents the initial fusion risk value, Ci represents the dynamic credibility of the data source corresponding to the i-th unified event data, and Pi represents the anomaly probability corresponding to the i-th unified event data.

[0062] Through the weighted fusion method described above, unified event data with higher dynamic credibility from the data source contributes more to the fusion computation, while unified event data with lower dynamic credibility from the data source contributes less to the fusion computation.

[0063] For example, if a temperature and humidity sensor has drifted multiple times in the past, even if the sensor outputs a high probability of anomalies, its effect on improving the initial fusion risk value is limited because the dynamic reliability of its corresponding data source is low. Conversely, if the switchgear temperature measuring device, partial discharge sensor, and video camera all have high data source dynamic reliability and all output high probabilities of anomalies, the initial fusion risk value will increase accordingly.

[0064] In multi-source fusion analysis, different unified event data may support each other or conflict with each other. For example, a humidity sensor may show extremely high humidity, but video images of the same area may not detect condensation, the cabinet temperature difference may be normal, and the partial discharge intensity may not change significantly, indicating a conflict. On the other hand, if the cable joint temperature rises while the load current rises and the ambient temperature remains stable, the unified event data support each other with a low degree of conflict.

[0065] Therefore, this embodiment generates a conflict penalty coefficient based on the degree of conflict between the unified event data in the candidate hidden danger events, and corrects the initial fusion risk value based on the conflict penalty coefficient to obtain the corrected fusion risk value.

[0066] Specifically, the degree of conflict between unified event data is calculated based on the indicator type, indicator value, anomaly probability, and dynamic reliability of the data source for each unified event data in the candidate potential event.

[0067] In one optional implementation, a rule base for indicator relationships can be pre-established. This rule base records the support or conflict relationships between different indicator types under different hazard types. For example, under the condensation hazard type, increased ambient humidity, abnormal temperature differences on the cabinet surface, suspected condensation in video, and enhanced partial discharge are support relationships; under the temperature rise hazard type, increased cable joint temperature and increased load current are support relationships; and under the smoke hazard type, smoke sensor triggering and video smoke recognition are support relationships.

[0068] If a certain unified event data has a high probability of being abnormal, while other unified event data with supporting relationships do not show any abnormalities, the conflict level is increased; if multiple unified event data with supporting relationships all show abnormalities, the conflict level is decreased.

[0069] A conflict penalty coefficient is generated based on the degree of conflict. The conflict penalty coefficient ranges from 0 to 1. The larger the conflict penalty coefficient, the more obvious the data conflict within the candidate potential event.

[0070] In one alternative implementation, the corrected fusion risk value is calculated as follows: R = R0·(1-K) Where R represents the corrected fusion risk value, R0 represents the initial fusion risk value, and K represents the conflict penalty coefficient.

[0071] When the conflict penalty coefficient is close to 0, it indicates that there is little conflict between the unified event data in the candidate hidden event, and the corrected fusion risk value is close to the initial fusion risk value; when the conflict penalty coefficient is large, it indicates that there is a large conflict between the unified event data in the candidate hidden event, and the corrected fusion risk value is significantly reduced relative to the initial fusion risk value.

[0072] By introducing a conflict penalty coefficient, this embodiment can reduce the probability of false alarms caused by isolated abnormal data.

[0073] After obtaining the corrected fusion risk value, the hazard level and hazard evidence chain are output based on the corrected fusion risk value.

[0074] In one optional implementation, multiple risk threshold ranges can be set. For example, when the corrected fusion risk value is less than the first risk threshold, a normal level is output; when the corrected fusion risk value is greater than or equal to the first risk threshold and less than the second risk threshold, a concern level is output; when the corrected fusion risk value is greater than or equal to the second risk threshold and less than the third risk threshold, a warning level is output; and when the corrected fusion risk value is greater than or equal to the third risk threshold, an alarm level is output.

[0075] The chain of evidence for potential hazards includes the hazard object, hazard type, triggering data source, triggering time, main evidence data, auxiliary evidence data, low-credibility evidence data, and handling recommendations.

[0076] Among them, primary evidence data can be unified event data that contributes significantly to the corrected fusion risk value and has high dynamic credibility of the data source; secondary evidence data can be unified event data that supports candidate potential events but has low dynamic credibility of the data source or low probability of anomalies; and low-credibility evidence data can be unified event data with high probability of anomalies but low dynamic credibility of the data source.

[0077] For example, regarding a condensation hazard in a switchgear, the output hazard evidence chain could include: the hazard object is switchgear No. 1, the hazard type is condensation hazard, the triggering data sources include temperature and humidity sensors, cabinet temperature measuring devices, partial discharge sensors, and video cameras, the triggering time is the corresponding corrected acquisition time range, the main evidence data is increased ambient humidity and abnormal cabinet temperature, the auxiliary evidence data is a slight increase in partial discharge intensity, the low-confidence evidence data is temperature and humidity sensor data with recent drift records, and the recommended actions are to start the dehumidification equipment, turn on the cabinet heating device, and arrange for maintenance personnel to review the data.

[0078] By outputting a chain of evidence for potential hazards, this embodiment can improve the interpretability of hazard analysis results, making it easier for maintenance personnel to quickly determine the source of hazards and the priority of handling them.

[0079] The historical consistency score and adaptive fusion time window are updated based on the feedback results.

[0080] Specifically, maintenance personnel conduct on-site handling or remote verification based on the hazard level and the chain of evidence, and then generate handling feedback results. These feedback results include reports of actual hazards, false alarms, and missed hazard reports.

[0081] When the feedback result is a genuine hazard feedback result, it indicates that the analysis result corresponding to the candidate hazard event is correct. This embodiment improves the historical consistency score of the data sources corresponding to the candidate hazard events. Furthermore, the historical consistency score of the data source corresponding to the primary evidence data can be improved, with a greater improvement than that of the data source corresponding to the auxiliary evidence data.

[0082] When the feedback result is a false alarm, it indicates that the candidate hazard event does not correspond to a real hazard. This embodiment reduces the historical consistency score of the data sources participating in the candidate hazard event. Furthermore, the historical consistency score of data sources corresponding to low-credibility evidence data can be reduced, or the historical consistency score of data sources with a high correlation to the false alarm result can be reduced.

[0083] When the feedback result is a missed report, it indicates that the corresponding hazard type failed to form a valid candidate hazard event in the previous fusion analysis. This embodiment adjusts the basic time window or preset evolution cycle parameter for the corresponding hazard type and regenerates an adaptive fusion time window. For example, if the condensation hazard type is missed multiple times, and on-site verification reveals that the time interval between anomalies in multiple data sources is long, the basic time window or preset evolution cycle parameter corresponding to the condensation hazard type can be increased. If the personnel boundary crossing hazard type is falsely aggregated, the basic time window corresponding to the personnel boundary crossing hazard type can be decreased.

[0084] Through the above-described feedback update method, this embodiment enables the dynamic reliability of data sources in different power distribution substations and the adaptive fusion time window to gradually adapt to local equipment status, communication conditions, and environmental characteristics, thereby improving the accuracy of fusion analysis during long-term operation.

[0085] Specific application example: Taking the analysis of condensation hazards in switchgear as an example. The temperature and humidity sensor near switchgear No. 1 in the substation continuously uploaded data on the increase in ambient humidity over a certain period of time. The cabinet temperature measuring device detected abnormal temperature differences on the cabinet surface, the partial discharge sensor detected a slight increase in the intensity of partial discharge, and the video camera identified suspected condensation reflections on the inner wall of the cabinet door.

[0086] The system first converts the above data into unified event data and calculates the dynamic reliability of the data source to which each unified event data belongs. Because the temperature and humidity sensor recently recorded an abnormal fluctuation, its data source dynamic reliability is lower than that of the cabinet temperature measuring device and the partial discharge sensor. Subsequently, the system corrects the acquisition time of the corresponding unified event data based on the historical transmission delay of each data source, thus obtaining the corrected acquisition time.

[0087] Since the hazard type is a condensation hazard, the system generates an adaptive fusion time window corresponding to the condensation hazard type based on the basic time window corresponding to the condensation hazard type, the maximum historical transmission delay among multiple data sources, and preset evolution cycle parameters. Within the adaptive fusion time window, the system aggregates the aforementioned unified event data into candidate hazard events according to the device object, spatial region, and condensation hazard type.

[0088] Next, the system calculates an initial fusion risk value based on the anomaly probability of each unified event data point and the dynamic reliability of the data source. Since the data source dynamic reliability of the cabinet temperature measurement device and the partial discharge sensor is high, and there is a supporting relationship between them and humidity increases, the initial fusion risk value is high. The system further generates a conflict penalty coefficient based on the degree of conflict between each unified event data point. Because video image data also supports condensation detection, the degree of conflict is low, the conflict penalty coefficient is small, and the resulting corrected fusion risk value remains at the warning level.

[0089] Finally, the system outputs the hazard level and hazard evidence chain, indicating that switchgear No. 1 has a risk of insulation damage due to condensation, and recommends starting the dehumidification equipment, turning on the internal heating device, and arranging for maintenance personnel to verify the issue. After on-site verification by maintenance personnel, condensation was confirmed to exist inside the cabinet, generating a true hazard feedback result. Based on this, the system improves the historical consistency score corresponding to the data source participating in this candidate hazard event.

[0090] This example demonstrates that the present invention does not simply superimpose multi-source heterogeneous data into alarms. Instead, it processes the dynamic reliability of the data sources, corrects the collection time, and adapts the fusion time window before fusion, and introduces a conflict penalty coefficient during the fusion process, thereby improving the accuracy and interpretability of the analysis of potential hazards in power distribution substations.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-source heterogeneous data fusion and analysis in intelligent operation and maintenance of power distribution substations, characterized in that, include: Acquire multi-source heterogeneous data generated by multiple data sources within the power distribution station, and convert the multi-source heterogeneous data into unified event data; Based on the data integrity score, timeliness score, fluctuation stability score, and historical consistency score corresponding to the unified event data, the dynamic reliability of the data source to which the unified event data belongs is calculated. Based on the reception time of the unified event data and the historical transmission delay of the data source to which the unified event data belongs, the collection time of the unified event data is corrected to obtain the corrected collection time of the unified event data. Based on the hazard type, the historical transmission delay, and the preset evolution cycle parameter, an adaptive fusion time window corresponding to the hazard type is generated; Based on the corrected acquisition time, the unified event data is aggregated according to the device object, spatial region and the type of hazard within the adaptive fusion time window to generate candidate hazard events; Based on the anomaly probability of each unified event data in the candidate potential event and the dynamic reliability of the data source, the initial fusion risk value of the candidate potential event is calculated; Based on the degree of conflict between the unified event data in the candidate hidden danger events, a conflict penalty coefficient is generated, and the initial fusion risk value is corrected according to the conflict penalty coefficient to obtain a corrected fusion risk value; The hazard level and hazard evidence chain are output based on the corrected fusion risk value, and the historical consistency score and the adaptive fusion time window are updated based on the handling feedback results.

2. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations according to claim 1, characterized in that, The multi-source heterogeneous data includes electrical operation data, equipment status data, environmental monitoring data, video image data, and access control work order data; The unified event data includes data source number, device object, spatial region, indicator type, indicator value, collection time, reception time, initial confidence level, and hazard type.

3. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations according to claim 2, characterized in that, The dynamic credibility of the data source is obtained by weighting the data integrity score, the timeliness score, the volatility stability score, and the historical consistency score, wherein the data integrity score, the timeliness score, the volatility stability score, and the historical consistency score are each configured with a credibility weight coefficient.

4. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of power distribution substations according to claim 3, characterized in that, The data integrity score is determined based on the missing proportion of the unified event data within a preset statistical period. The timeliness score is determined based on the historical transmission delay. The fluctuation stability score is determined based on the number of abnormal jumps in the indicator value within a preset statistical period. The historical consistency score is determined based on the number of times the candidate potential hazard events and the handling feedback results are consistent with each other, as indicated by the data source.

5. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 1, characterized in that, The steps for obtaining the corrected acquisition time include: Obtain multiple transmission delay samples from the same data source within a historical statistical period; The historical transmission delay of the data source is calculated based on multiple transmission delay samples; The corrected acquisition time is obtained by subtracting the historical transmission delay from the reception time.

6. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 1, characterized in that, The adaptive fusion time window is determined based on the base time window corresponding to the hazard type, the maximum historical transmission delay among the multiple data sources associated with the hazard type, and the preset evolution cycle parameter, and the maximum historical transmission delay and the preset evolution cycle parameter are weighted and corrected by the time window adjustment coefficient.

7. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 1, characterized in that, The steps for generating the candidate potential event include: Multiple unified event data that are the same in terms of equipment object, spatial region, and hazard type, and whose correction collection time falls within the same adaptive fusion time window, are grouped into the same event set. When the number of unified event data in the event set reaches a preset data quantity threshold, or when the data source type corresponding to the unified event data in the event set reaches a preset data source type threshold, the event set is determined as the candidate potential event.

8. The multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 1, characterized in that, The initial fusion risk value is obtained by weighting the anomaly probability corresponding to each unified event data in the candidate hidden danger events with the dynamic credibility of the data source, wherein the dynamic credibility of the data source is used to determine the contribution weight of the corresponding anomaly probability in the fusion calculation.

9. A multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 8, characterized in that, The corrected fusion risk value is obtained by attenuating and correcting the initial fusion risk value using the conflict penalty coefficient. The larger the conflict penalty coefficient, the greater the attenuation of the corrected fusion risk value relative to the initial fusion risk value.

10. A multi-source heterogeneous data fusion and analysis method for intelligent operation and maintenance of substations according to claim 6, characterized in that, The chain of evidence for potential hazards includes the hazard object, hazard type, triggering data source, triggering time, main evidence data, auxiliary evidence data, low-credibility evidence data, and handling recommendations; The feedback results include feedback results on actual potential hazards, feedback results on false alarms, and feedback results on missed alarms. When the handling feedback result is the actual hidden danger feedback result, the historical consistency score corresponding to the data source participating in the candidate hidden danger event is increased; When the handling feedback result is the false alarm feedback result, the historical consistency score corresponding to the data source participating in the candidate hidden danger event is reduced; When the handling feedback result is the missed report feedback result, adjust the basic time window or the preset evolution cycle parameter corresponding to the hidden danger type, and regenerate the adaptive fusion time window.

Images