Transformer intelligent monitoring and fault diagnosis system and method based on multi-source data fusion

Abstract

The application is suitable for the technical field of power equipment monitoring, and provides a transformer intelligent monitoring and fault diagnosis system and method based on multi-source data fusion, wherein the method generates a difference mode benchmark library according to seasons and load rate dimensions by integrating historical working condition data; collects real-time data and compares the real-time data with benchmark data to output abnormal information; constructs an association graph based on historical interaction data, and calls monitoring data of adjacent equipment in the same dimension; compares the associated data with the benchmark data to generate a fault report or optimize the benchmark library. Through multi-dimensional benchmark setting, associated equipment collaborative analysis and dynamic optimization mechanism, the application improves the accuracy and adaptability of fault diagnosis, and realizes intelligent monitoring and diagnosis of the transformer.

Description

Technical Field

[0001] This invention belongs to the field of power equipment monitoring technology, and in particular relates to a transformer intelligent monitoring and fault diagnosis system and method based on multi-source data fusion. Background Technology

[0002] As a core piece of equipment in the power system, the operating status of transformers directly affects the safety and stability of the power grid. With the development of smart grid technology, transformer monitoring and fault diagnosis have evolved from traditional manual inspections and single-parameter monitoring to intelligent, multi-source data-driven approaches. Currently, multi-source data fusion technology is widely used in equipment condition assessment. By integrating historical operating data, real-time monitoring parameters, and environmental data, it improves the timeliness and accuracy of fault diagnosis, becoming an important development trend in the intelligent management of power equipment.

[0003] However, existing technologies for transformer monitoring have the following problems: the dimensional division of the benchmark library is too simplistic and does not fully consider the cross-influence of seasons and load rates, resulting in unclear definitions of normal parameter ranges under different operating conditions and easy misjudgments; equipment correlation analysis relies only on simple connection relationships and does not quantify the correlation strength, making it difficult to accurately locate fault propagation paths; the benchmark library lacks a dynamic optimization mechanism and cannot adjust dimensional weights according to misjudgment cases in actual operation, resulting in insufficient adaptability and difficulty in meeting the intelligent diagnostic needs under complex operating conditions. Summary of the Invention

[0004] The purpose of this invention is to provide a transformer intelligent monitoring and fault diagnosis system and method based on multi-source data fusion, aiming to solve the technical problems existing in the prior art as identified in the background art.

[0005] This invention is implemented as follows: a transformer intelligent monitoring and fault diagnosis system based on multi-source data fusion, the system comprising:

[0006] The differential pattern benchmark library generation module is used to integrate historical transformer operating condition data, divide the data into subsets according to seasonal and load factor dimensions, extract the normal fluctuation range of monitoring parameters under each dimension combination, and generate the differential pattern benchmark library.

[0007] An anomaly detection module is used to collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference pattern benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold.

[0008] The association graph construction module is used to construct a substation equipment association graph based on the historical operation interaction data of transformers. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, it locates the set of adjacent equipment and retrieves the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred as the association monitoring data.

[0009] The fault location and optimization module is used to convert the associated monitoring data into associated data vectors, compare the deviation of the associated data vectors with the corresponding dimension of the benchmark data in the difference pattern benchmark library, generate a fault location report if an associated anomaly occurs in the set of adjacent devices, otherwise start the dimension weight optimization of the difference pattern benchmark library.

[0010] Another objective of this invention is to provide a method for intelligent monitoring and fault diagnosis of transformers based on multi-source data fusion, the method comprising:

[0011] Integrate historical operating data of transformers, divide the data into subsets according to seasonal and load factor dimensions, extract the normal fluctuation range of monitoring parameters under each dimension combination, and generate a benchmark library of difference patterns.

[0012] Collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference pattern benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold.

[0013] Based on historical operation interaction data of transformers, a substation equipment association map is constructed. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, the set of adjacent equipment is located, and the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred is retrieved as the associated monitoring data.

[0014] The associated monitoring data is converted into an associated data vector. The deviation of the associated data vector from the benchmark data of the corresponding dimension in the difference pattern benchmark library is compared. If an associated anomaly occurs in the set of adjacent devices, a fault location report is generated; otherwise, the dimension weight optimization of the difference pattern benchmark library is initiated.

[0015] As a further aspect of the present invention, the generation of the difference pattern benchmark library specifically includes:

[0016] The integrated historical operating condition data of the transformer is preprocessed, and the historical operating condition data of the transformer includes historical fault data and historical monitoring parameters;

[0017] For historical monitoring parameters, seasonal and load factor dimensions are defined separately, and then cross-combined to generate seasonal and load factor data subsets.

[0018] For monitoring parameters in the seasonal and load factor data subsets, calculate the normal fluctuation range, perform reverse verification by combining historical fault data, automatically correct the benchmark fluctuation threshold under extreme operating conditions, and generate a benchmark library of difference patterns with dimension labels.

[0019] As a further aspect of the present invention, the step of outputting an abnormal device identifier and abnormal characteristic information when the deviation exceeds a set deviation threshold specifically includes:

[0020] The real-time monitoring parameters collected in real time are aggregated into time series data according to a preset time interval and converted into a real-time data vector consistent with the format of the differential pattern benchmark library.

[0021] Extract the seasonal and load factor labels from the real-time data vectors, match them with the corresponding dimension of the benchmark data in the difference pattern benchmark library, and calculate the deviation.

[0022] An anomaly determination threshold is set based on historical fault data. The calculated deviation is compared with the anomaly determination threshold. If the deviation exceeds the anomaly determination threshold, it is determined as an initial anomaly. Anomaly feature information including transformer ID, anomaly monitoring parameters, and occurrence time is generated and output.

[0023] As a further aspect of the present invention, the benchmark data of the corresponding dimension in the matching difference pattern benchmark library and the deviation degree are calculated, specifically as follows:

[0024] ;

[0025] in, This indicates the deviation of the real-time data vector from the baseline data. The first one in the real-time data vector of the target transformer Standardized values ​​of each monitoring parameter It is the first time that the corresponding seasonal and load factor dimensions in the difference pattern benchmark library are used to determine the difference pattern. The baseline average of each monitoring parameter The number of parameters to be monitored.

[0026] As a further aspect of the present invention, the construction of the substation equipment association map specifically includes:

[0027] To address the initial anomalies, based on historical transformer operating data, interaction data between transformers is extracted, and a substation equipment association graph is constructed with transformers as nodes and association strength as edge weights.

[0028] Centered on the node corresponding to the abnormal equipment identifier, the top 5 nodes with the highest edge weights in the substation equipment association graph are selected as the set of adjacent transformers, and the association type of each transformer is marked.

[0029] Based on the timestamp in the abnormal feature information, the seasonal dimension and load rate dimension of the transformer anomaly at the time of occurrence are identified and determined as the initial anomaly. The historical monitoring parameters and real-time monitoring parameters of adjacent transformer sets under the same dimension are automatically retrieved and merged into associated monitoring data.

[0030] As a further embodiment of the present invention, the correlation strength is calculated using the following formula:

[0031] ;

[0032] in, Indicates transformer With transformer The strength of the association between them, that is, the weight of the edges in the association graph. For transformer With transformer The electrical connection factor is set to 1 for direct shared busbars, 0.5 for indirect connections, and 0 for no electrical connection. It is a transformer With transformer The physical distance between them This refers to the frequency of historical fault co-occurrence, i.e., the frequency at which the transformer experienced faults in the past. With transformer The ratio of the number of times an anomaly occurs to the total number of failures. These are the weighting coefficients;

[0033] The association graph is represented as follows:

[0034] ;

[0035] for The matrix, This represents the total number of transformers within the substation. The first in the matrix Line number The elements of the column represent transformers. With transformer The strength of the correlation between them.

[0036] As a further aspect of the present invention, the comparison of the deviation between the associated data vector and the corresponding dimension of the benchmark data in the difference pattern benchmark library specifically includes:

[0037] The associated monitoring data is converted into associated data vectors, and the deviation between the associated data vectors of each adjacent transformer and the corresponding dimension benchmark data is calculated.

[0038] If more than 30% of the transformers in an adjacent transformer set are identified as having deviations exceeding the anomaly judgment threshold, they are defined as associated anomalies. The fault propagation path of the transformers identified as the initial anomalies is determined by combining the edge weights of the association graph, and a fault location report containing the fault source and the scope of impact is automatically generated.

[0039] If no associated anomalies are identified among adjacent transformers, the judgment result of the transformer identified as the initial anomaly is defined as a misjudgment. The contribution of the seasonal dimension and load factor dimension to this misjudgment is analyzed, and key influencing dimensions are identified. Based on the contribution, the weight value of key influencing dimensions is increased proportionally, and the benchmark fluctuation thresholds of the seasonal dimension data subset and the load factor dimension data subset in the difference pattern benchmark library are retrained.

[0040] As a further aspect of the present invention, the contribution of the seasonal dimension and load factor dimension to the misjudgment is specifically as follows:

[0041] ;

[0042] ;

[0043] in, The contribution of the seasonal dimension to this misjudgment. The contribution of the load factor to this misjudgment. , This refers to the deviation of the target transformer in the current seasonal context during this misjudgment. This represents the average deviation from the historical normal state under this seasonal dimension. This refers to the deviation of the target transformer from its current load factor in this misjudgment. This represents the average deviation from historical normal conditions under this load factor dimension. Indicates the first Deviation of the target transformer in each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, That is, the deviation in the load factor dimension. Indicates the first The average deviation of the historical normal state under each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, ; The total number of dimensions participating in the contribution analysis. That is, the seasonal dimension and the load factor dimension.

[0044] The beneficial effects of this invention are:

[0045] This invention achieves precision, systematization, and adaptability in intelligent transformer monitoring and fault diagnosis through multi-source data fusion technology. The differential pattern benchmark library, combined with the cross-dimensional dimensions of season and load rate, divides data subsets, making the definition of normal fluctuation ranges more closely aligned with actual operating scenarios and significantly reducing misjudgments caused by differences in operating conditions. The correlation graph construction module, by quantifying the correlation strength of devices, can accurately locate adjacent devices related to abnormal devices. Combined with the comparison of correlation data in the same dimension, it enables the tracing of fault propagation paths and the determination of the scope of impact. The fault location and optimization module generates accurate fault reports through correlation anomaly judgment and dynamically optimizes the dimensional weights of the benchmark library using misjudgment cases, enabling the system to continuously adapt to changes in operating conditions and form a closed loop of monitoring, diagnosis, and optimization. The overall solution fully integrates multi-source information such as historical data, real-time data, and related device data, significantly improving the accuracy and timeliness of fault diagnosis and providing a reliable guarantee for the safe and stable operation of the power grid. Attached Figure Description

[0046] Figure 1 This is a structural block diagram of a transformer intelligent monitoring and fault diagnosis system based on multi-source data fusion provided in an embodiment of the present invention.

[0047] Figure 2 A flowchart of a transformer intelligent monitoring and fault diagnosis method based on multi-source data fusion provided in an embodiment of the present invention;

[0048] Figure 3 A flowchart for generating a differential pattern benchmark library provided in an embodiment of the present invention;

[0049] Figure 4 This is a flowchart provided by an embodiment of the present invention, showing how to output an abnormal device identifier and abnormal feature information when the deviation exceeds a set deviation threshold;

[0050] Figure 5 This is a flowchart for constructing a substation equipment association map provided in an embodiment of the present invention;

[0051] Figure 6 This is a flowchart illustrating the deviation state between the associated data vector and the corresponding dimension of the benchmark data in the difference pattern benchmark library, provided as an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] Figure 1 The structural block diagram of the transformer intelligent monitoring and fault diagnosis system based on multi-source data fusion provided in the embodiments of the present invention is as follows: Figure 1 As shown, the system includes:

[0054] The differential pattern benchmark library generation module 100 is used to integrate historical transformer operating condition data, divide the data subsets according to seasonal and load rate dimensions, extract the normal fluctuation range of monitoring parameters under each dimension combination, and generate the differential pattern benchmark library.

[0055] Anomaly detection module 200 is used to collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference pattern benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold.

[0056] The association graph construction module 300 is used to construct a substation equipment association graph based on the historical operation interaction data of transformers. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, it locates the set of adjacent equipment and retrieves the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred as the association monitoring data.

[0057] The fault location and optimization module 400 is used to convert the associated monitoring data into an associated data vector, compare the deviation of the associated data vector with the benchmark data of the corresponding dimension in the difference pattern benchmark library, generate a fault location report if an associated anomaly occurs in the set of adjacent devices, otherwise start the dimension weight optimization of the difference pattern benchmark library.

[0058] Figure 2 The flowchart of the intelligent monitoring and fault diagnosis method for transformers based on multi-source data fusion provided in the embodiments of the present invention is as follows: Figure 2 As shown, the method includes:

[0059] S100 integrates historical transformer operating data, divides data subsets by seasonal and load factor dimensions, extracts the normal fluctuation range of monitoring parameters under each dimension combination, and generates a differential pattern benchmark library.

[0060] The integrated transformer historical operating data includes historical fault data accumulated over a long period of time and various monitoring parameters, such as temperature, pressure, insulation resistance, current, and voltage. The reason for integrating this multi-source data is that the operating status of the transformer is affected by a variety of factors, and a single type of data cannot fully reflect its normal fluctuation characteristics. Only by integrating multi-dimensional data can sufficient basis be provided for the subsequent determination of the benchmark.

[0061] The data is divided into subsets based on seasonality and load factor, and then cross-combined to consider the operating environment and load characteristics of transformers. The seasonal dimension is divided into spring, summer, autumn, and winter. Significant differences in temperature, humidity, and other environmental factors across different seasons or quarters directly affect the transformer's heat dissipation efficiency and insulation performance. In summer, the normal temperature range of a transformer is higher than in winter due to the high temperatures. The load factor dimension is divided into low, medium, and high loads based on electricity demand. The fluctuation ranges of electrical parameters such as current and voltage of the transformer differ significantly under different loads; the normal current range under high load is usually greater than under low load. Cross-combining these two dimensions, such as high load in summer and low load in winter, allows for a more detailed correspondence to the actual operating scenarios of transformers, avoiding the ambiguity caused by a single-dimensional division. This is because the normal parameter ranges of transformers often differ under the same load factor in different seasons, and even under different load factors within the same season.

[0062] After segmenting the data, the normal fluctuation range of monitoring parameters under each dimension combination is extracted. Statistical analysis is used to calculate the reasonable range for each parameter. Reverse verification using historical fault data is crucial for improving the reliability of the benchmark library. Historical fault data records the parameter states when transformer faults occur. By comparing this data with the initially calculated normal range, it can be identified whether the normal range is defined too broadly or too narrowly. If a parameter repeatedly exceeds the initially set normal range in historical faults, it indicates that the range may need adjustment. Furthermore, for extreme operating conditions, such as high-load periods during the summer with sustained high temperatures or low-load periods during the winter with extreme cold, transformer parameter fluctuations may exceed the normal range. In these cases, the benchmark fluctuation threshold needs to be automatically corrected to ensure the benchmark library covers these special scenarios and avoids misjudging normal fluctuations under extreme conditions as abnormalities. The final generated benchmark library with dimension-labeled difference patterns provides clear normal parameter reference standards for each specific operating condition.

[0063] like Figure 3 As shown, the generation of the difference pattern benchmark library specifically includes:

[0064] S110, preprocess the integrated transformer historical operating condition data, which includes historical fault data and historical monitoring parameters;

[0065] S120 divides historical monitoring parameters into seasonal and load factor dimensions, and cross-combines them to generate seasonal and load factor data subsets.

[0066] S130 calculates the normal fluctuation range for monitoring parameters in the seasonal and load factor data subsets, performs reverse verification by combining historical fault data, automatically corrects the benchmark fluctuation threshold under extreme operating conditions, and generates a benchmark library of difference patterns with dimension labels.

[0067] S200: Collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference mode benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold.

[0068] The real-time monitoring data collection covers various key parameters during transformer operation. These parameters reflect the physical state and electrical characteristics of the equipment in real time. Continuous collection ensures immediate perception of changes in equipment status, preventing the nascent stages of faults from being overlooked due to data lag. Aggregating real-time monitoring parameters into time-series data at preset time intervals is not a simple data accumulation, but rather a process of integrating data over time to filter out transient interference signals and highlight the trend characteristics of parameter changes. For example, short-term voltage fluctuations may be occasional interference, but fluctuations over multiple consecutive time intervals may indicate potential problems. Aggregation processing makes the data more valuable for analysis.

[0069] Time-series data is converted into real-time data vectors consistent with the format of the differential model benchmark library, eliminating differences in the units and numerical ranges of various parameters and providing a basis for direct comparison between real-time data and benchmark data. Since the data in the benchmark library is structured and stored according to seasonal and load rate dimensions, the unified data vector format ensures smooth subsequent matching and calculation, avoiding comparison errors caused by format incompatibility. Extracting the seasonal and load rate labels from the real-time data vectors and matching them with the corresponding dimension-based benchmark data continues the logic of constructing benchmarks by dimension in S100. This is because the normal operating parameters of transformers naturally differ under different seasons and load rates. For example, the reduced heat dissipation efficiency in high-temperature seasons will result in a higher normal temperature range than in low-temperature seasons, and the normal current range under high load will also be greater than under low load.

[0070] Calculating the deviation between real-time data vectors and baseline data essentially involves using quantitative analysis to measure the difference between the current state and the normal state. This quantitative approach is more objective than qualitative judgment and can effectively avoid subjective biases from human experience. Setting anomaly judgment thresholds based on historical fault data incorporates the abnormal characteristics presented in past fault cases into the judgment criteria. The threshold settings are empirical values ​​that align with the actual fault patterns of the equipment, ensuring that genuine abnormal states are captured promptly while reducing the misjudgment of normal fluctuations as abnormalities.

[0071] When the deviation exceeds the threshold, the output includes abnormal feature information such as transformer ID, abnormal monitoring parameters, and occurrence time, which clarifies the object and specific parameters of the abnormality. It also provides a time anchor for subsequent correlation graph analysis, ensuring that subsequent steps can accurately retrieve data of related equipment in the same time dimension, forming a closed loop from abnormality identification to in-depth diagnosis.

[0072] like Figure 4 As shown, the step of outputting an abnormal device identifier and abnormal characteristic information when the deviation exceeds a set deviation threshold specifically includes:

[0073] S210 aggregates the real-time monitoring parameters collected in real time into time series data according to a preset time interval, and converts them into real-time data vectors consistent with the format of the differential mode benchmark library.

[0074] S220: Extract the seasonal label and load factor label of the real-time data vector, match it with the benchmark data of the corresponding dimension in the difference pattern benchmark library, and calculate the deviation.

[0075] S230 sets an anomaly judgment threshold based on historical fault data, compares the calculated deviation with the anomaly judgment threshold, and if it exceeds the anomaly judgment threshold, it is judged as an initial anomaly, and generates and outputs anomaly feature information including transformer ID, anomaly monitoring parameters, and occurrence time.

[0076] In this step, the benchmark data for the corresponding dimension in the matching difference pattern benchmark library is used, and the deviation is calculated, specifically as follows:

[0077] ;

[0078] in, This indicates the deviation of the real-time data vector from the baseline data. The first one in the real-time data vector of the target transformer Standardized values ​​of each monitoring parameter It is the first time that the corresponding seasonal and load factor dimensions in the difference pattern benchmark library are used to determine the difference pattern. The baseline average of each monitoring parameter The number of parameters to be monitored.

[0079] S300: Based on the historical operation interaction data of transformers, a substation equipment association map is constructed. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, the set of adjacent equipment is located, and the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred is retrieved as the associated monitoring data.

[0080] A substation equipment correlation map is constructed based on historical transformer operation interaction data to uncover potential mutual influence relationships between equipment. This interaction data includes not only electrical connections between equipment, such as whether they share a busbar and the connection method, but also physical layout information such as the distance between equipment and the frequency of simultaneous anomalies in historical faults. The reason for integrating this multi-dimensional interaction data is that the operating status of a transformer is not isolated; its anomalies may originate from fault propagation from adjacent equipment or affect related equipment. Equipment with close electrical connections is more likely to transmit anomalies through circuits, equipment with close physical distances may be affected by the same environmental factors, and equipment that has experienced multiple simultaneous faults in the past has a higher fault correlation. By quantifying these relationships into correlation strength (i.e., edge weights), the correlation map can intuitively present the degree of influence between equipment, providing a basis for subsequent location of key related equipment.

[0081] Starting with the transformer corresponding to the abnormal equipment identifier, the set of adjacent equipment is located. Nodes with higher edge weights are selected, focusing on the equipment most closely associated with the abnormal equipment. Since equipment with strong association is more likely to interact directly or indirectly with the abnormal equipment, its operating status often has a higher correlation with the abnormal characteristics of the abnormal equipment. Prioritizing the analysis of these devices reduces interference from irrelevant data and improves analysis efficiency. Retrieving monitoring data from these adjacent devices under the same seasonal and load factor dimensions as when the abnormal characteristics occurred is to ensure that the associated data and the abnormal data are within the same operating condition context. As mentioned earlier, season and load factor significantly affect transformer parameters; the normal fluctuation range of the same equipment differs under different seasons or load factors. If the operating condition dimensions of the associated data and the abnormal data are inconsistent, the comparison results will lose their reference value. Therefore, associated data with a unified dimension ensures comparability with the abnormal data.

[0082] By constructing a correlation graph, scattered device data is linked together through relationships, providing a broader analytical context for previously isolated anomaly information. Locating adjacent device sets and retrieving related data of the same dimension ensures the correlation and comparability of multi-device data, avoiding analytical biases caused by differences in data background. This multi-device collaborative analysis model can more comprehensively determine the nature of anomalies—whether they are due to device malfunctions or the influence of related devices—providing richer clues for subsequent fault localization. This expands intelligent diagnosis based on multi-source data fusion beyond single-device parameter fluctuations to include multi-device state collaboration, significantly improving the depth and accuracy of fault diagnosis.

[0083] like Figure 5 As shown, the construction of the substation equipment association map specifically includes:

[0084] S310, for the existence of initial anomalies, based on the historical operating data of transformers, extract the interaction data between transformers and construct a substation equipment association map with transformers as nodes and association strength as edge weights;

[0085] S320: Taking the node corresponding to the abnormal equipment identifier as the center, select the top 5 nodes with the highest edge weight in the substation equipment association graph as the set of adjacent transformers, and mark the association type of each transformer.

[0086] S330 identifies the seasonal dimension and load rate dimension of the transformer anomaly when it occurs, based on the timestamp in the anomaly feature information. It then automatically retrieves the historical and real-time monitoring parameters of adjacent transformer sets under the same dimension and merges them into associated monitoring data.

[0087] In this step, the correlation strength is calculated using the following formula:

[0088] ;

[0089] in, Indicates transformer With transformer The strength of the association between them, that is, the weight of the edges in the association graph. For transformer With transformer The electrical connection factor is set to 1 for direct shared busbars, 0.5 for indirect connections, and 0 for no electrical connection. It is a transformer With transformer The physical distance between them This refers to the frequency of historical fault co-occurrence, i.e., the frequency at which the transformer experienced faults in the past. With transformer The ratio of the number of times an anomaly occurs to the total number of failures. These are the weighting coefficients;

[0090] The association graph is represented as follows:

[0091] ;

[0092] for The matrix, This represents the total number of transformers within the substation. The first in the matrix Line number The elements of the column represent transformers. With transformer The strength of the correlation between them.

[0093] S400, the associated monitoring data is converted into an associated data vector, and the deviation of the associated data vector from the benchmark data of the corresponding dimension in the difference pattern benchmark library is compared. If an associated anomaly occurs in the set of adjacent devices, a fault location report is generated; otherwise, the dimension weight optimization of the difference pattern benchmark library is initiated.

[0094] The associated monitoring data is converted into associated data vectors to ensure that the monitoring data of associated equipment is consistent in format with the difference pattern benchmark library. Only when the data format is unified can the monitoring parameters of different equipment be quantitatively evaluated under the same analytical framework, avoiding comparison bias caused by differences in data format. Subsequently, the deviation of the associated data vectors of each adjacent transformer from the corresponding dimension benchmark data is calculated. The purpose is to clarify the degree of difference between the associated equipment and the normal state through quantitative indicators. This quantitative analysis can more objectively reflect the severity of the anomaly than qualitative judgment, providing a measurable basis for subsequent judgment of associated anomalies.

[0095] The approach to determining whether an anomaly is associated with a set of adjacent devices is based on the principle of fault propagation: as a node in a substation network, a transformer's fault often affects associated devices through electrical connections, physical environment, and other paths. If most associated devices simultaneously show deviations exceeding a threshold, it indicates that the anomaly is not isolated but has a propagating effect, likely originating from a fault propagation source. Conversely, if there are no obvious anomalies in the associated devices, it suggests that the initial anomaly determination may be affected by the deviation in the dimensional definition of the benchmark library.

[0096] Once the correlation anomaly is confirmed, the fault propagation path is determined by combining the edge weights of the correlation graph. This is because the edge weights directly reflect the correlation strength between devices, and paths with high correlation strength are more likely to be the main channels for fault propagation. Based on this, the source of the fault can be traced and the scope of impact can be determined. The final fault location report integrates multi-source data from abnormal devices and related devices, realizing a panoramic diagnosis from a single anomaly point to the system fault network.

[0097] If no correlation anomalies are identified, the initial anomaly is judged as a misjudgment, and benchmark library optimization is initiated. Analyzing the contribution of seasonal and load factor dimensions to misjudgments is to accurately locate dimensions with biases in the benchmark library. The weight of different dimensions influencing equipment parameter fluctuations is not fixed. If a certain dimension contributes highly to misjudgments, it indicates that the benchmark fluctuation threshold for that dimension does not accurately reflect the actual operating conditions. The weights need to be adjusted according to the contribution ratio, and the benchmark library needs to be retrained. This adjustment is not random but based on multi-source data feedback from real-time misjudgment cases, enabling the benchmark library to dynamically adapt to changes in the equipment operating environment and load, reducing the possibility of similar misjudgments in the future.

[0098] like Figure 6As shown, the deviation status of comparing the associated data vector with the corresponding dimension of the benchmark data in the difference pattern benchmark library specifically includes:

[0099] S410, convert the associated monitoring data into associated data vectors, and calculate the deviation of each adjacent transformer associated data vector from the corresponding dimension benchmark data;

[0100] S420: If more than 30% of the transformers in the adjacent transformer set are identified as having a deviation exceeding the anomaly judgment threshold, they are defined as associated anomalies. The fault propagation path of the transformers identified as the initial anomalies is determined by combining the edge weights of the association graph, and a fault location report containing the fault source and the scope of influence is automatically generated.

[0101] S430 If no associated anomaly is identified among adjacent transformers, the judgment result of the transformer that was initially identified as an anomaly is defined as a misjudgment. The contribution of the seasonal dimension and the load factor dimension to this misjudgment is analyzed, and key influencing dimensions are identified. The weight value of key influencing dimensions is increased proportionally according to the contribution. The benchmark fluctuation thresholds of the seasonal dimension data subset and the load factor dimension data subset in the difference pattern benchmark library are retrained.

[0102] In this step, the contribution of the seasonal dimension and load factor dimension to the misjudgment is specifically as follows:

[0103] ;

[0104] ;

[0105] in, The contribution of the seasonal dimension to this misjudgment. The contribution of the load factor to this misjudgment. , This refers to the deviation of the target transformer in the current seasonal context during this misjudgment. This represents the average deviation from the historical normal state under this seasonal dimension. This refers to the deviation of the target transformer from its current load factor in this misjudgment. This represents the average deviation from historical normal conditions under this load factor dimension. Indicates the first Deviation of the target transformer in each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, That is, the deviation in the load factor dimension. Indicates the first The average deviation of the historical normal state under each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, ; The total number of dimensions participating in the contribution analysis. That is, the seasonal dimension and the load factor dimension.

[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transformer intelligent monitoring and fault diagnosis system based on multi-source data fusion, characterized in that, The system includes: The differential pattern benchmark library generation module is used to integrate historical transformer operating condition data, divide the data into subsets according to seasonal and load factor dimensions, extract the normal fluctuation range of monitoring parameters under each dimension combination, and generate the differential pattern benchmark library. An anomaly detection module is used to collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference pattern benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold. The association graph construction module is used to construct a substation equipment association graph based on the historical operation interaction data of transformers. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, it locates the set of adjacent equipment and retrieves the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred as the association monitoring data. The fault location and optimization module is used to convert the associated monitoring data into associated data vectors, compare the deviation of the associated data vectors with the benchmark data of the corresponding dimension in the difference pattern benchmark library, generate a fault location report if an association anomaly occurs in the set of adjacent devices, otherwise start the dimension weight optimization of the difference pattern benchmark library. The generation of the difference pattern benchmark library specifically includes: The integrated historical operating condition data of the transformer is preprocessed, and the historical operating condition data of the transformer includes historical fault data and historical monitoring parameters; For historical monitoring parameters, seasonal and load factor dimensions are defined separately, and then cross-combined to generate seasonal and load factor data subsets. For monitoring parameters in the seasonal and load factor data subsets, calculate the normal fluctuation range, perform reverse verification by combining historical fault data, automatically correct the benchmark fluctuation threshold under extreme operating conditions, and generate a benchmark library of difference patterns with dimension labels. The construction of the substation equipment association map specifically includes: To address the initial anomalies, based on historical transformer operating data, interaction data between transformers is extracted, and a substation equipment association graph is constructed with transformers as nodes and association strength as edge weights. Centered on the node corresponding to the abnormal equipment identifier, the top 5 nodes with the highest edge weights in the substation equipment association graph are selected as the set of adjacent transformers, and the association type of each transformer is marked. Based on the timestamp in the abnormal feature information, the seasonal dimension and load rate dimension of the transformer anomaly at the time of the initial anomaly are identified and determined. The historical and real-time monitoring parameters of adjacent transformer sets under the same dimension are automatically retrieved and merged into associated monitoring data. The correlation strength is calculated using the following formula: ； in, Indicates transformer With transformer The strength of the association between them, that is, the weight of the edges in the association graph. For transformer With transformer The electrical connection factor is set to 1 for direct shared busbars, 0.5 for indirect connections, and 0 for no electrical connection. It is a transformer With transformer The physical distance between them This refers to the frequency of historical fault co-occurrence, i.e., the frequency at which the transformer experienced faults in the past. With transformer The ratio of the number of times an anomaly occurs to the total number of failures. These are the weighting coefficients; The association graph is represented as follows: ； for The matrix, This represents the total number of transformers within the substation. The first in the matrix Line number The elements of the column represent transformers. With transformer The strength of the correlation between them; The comparison of the deviation between the associated data vector and the corresponding dimension of the benchmark data in the difference pattern benchmark library specifically includes: The associated monitoring data is converted into associated data vectors, and the deviation between the associated data vectors of each adjacent transformer and the corresponding dimension benchmark data is calculated. If more than 30% of the transformers in an adjacent transformer set are identified as having deviations exceeding the anomaly judgment threshold, they are defined as associated anomalies. The fault propagation path of the transformers identified as the initial anomalies is determined by combining the edge weights of the association graph, and a fault location report containing the fault source and the scope of impact is automatically generated. If no associated anomalies are identified among adjacent transformers, the judgment result of the transformer identified as the initial anomaly is defined as a misjudgment. The contribution of the seasonal dimension and load factor dimension to this misjudgment is analyzed, and key influencing dimensions are identified. Based on the contribution, the weight value of key influencing dimensions is increased proportionally, and the benchmark fluctuation thresholds of the seasonal dimension data subset and the load factor dimension data subset in the difference pattern benchmark library are retrained.

2. The system according to claim 1, characterized in that, The method for implementing the transformer intelligent monitoring and fault diagnosis system based on multi-source data fusion includes: Integrate historical operating data of transformers, divide the data into subsets according to seasonal and load factor dimensions, extract the normal fluctuation range of monitoring parameters under each dimension combination, and generate a benchmark library of difference patterns. Collect real-time monitoring data of the target transformer, convert the real-time monitoring data into a real-time data vector, match it with the benchmark data in the difference pattern benchmark library corresponding to the seasonal dimension and load rate dimension, calculate the deviation between the real-time data vector and the benchmark data, and output the abnormal equipment identifier and abnormal feature information when the deviation exceeds the set deviation threshold. Based on historical operation interaction data of transformers, a substation equipment association map is constructed. Taking the transformer corresponding to the abnormal equipment identifier as the starting node, the set of adjacent equipment is located, and the monitoring data of the set of adjacent equipment under the same seasonal dimension and load rate dimension as when the abnormal feature information occurred is retrieved as the associated monitoring data. The associated monitoring data is converted into an associated data vector. The deviation of the associated data vector from the benchmark data of the corresponding dimension in the difference pattern benchmark library is compared. If an associated anomaly occurs in the set of adjacent devices, a fault location report is generated; otherwise, the dimension weight optimization of the difference pattern benchmark library is initiated.

3. The system according to claim 1, characterized in that, When the deviation exceeds a set deviation threshold, the abnormal device identifier and abnormal characteristic information are output, specifically including: The real-time monitoring parameters collected in real time are aggregated into time series data according to a preset time interval and converted into a real-time data vector consistent with the format of the differential pattern benchmark library. Extract the seasonal and load factor labels from the real-time data vectors, match them with the corresponding dimension of the benchmark data in the difference pattern benchmark library, and calculate the deviation. An anomaly determination threshold is set based on historical fault data. The calculated deviation is compared with the anomaly determination threshold. If the deviation exceeds the anomaly determination threshold, it is determined as an initial anomaly. Anomaly feature information including transformer ID, anomaly monitoring parameters, and occurrence time is generated and output.

4. The system according to claim 3, characterized in that, The benchmark data for the corresponding dimension in the matching difference pattern benchmark library is used to calculate the deviation, specifically as follows: ； in, This indicates the deviation of the real-time data vector from the baseline data. The first one in the real-time data vector of the target transformer Standardized values ​​of each monitoring parameter It is the first time that the corresponding seasonal and load factor dimensions in the difference pattern benchmark library are used to determine the difference pattern. The baseline average of each monitoring parameter The number of parameters to be monitored.

5. The system according to claim 1, characterized in that, The contribution of the seasonal dimension and load factor dimension to this misjudgment is analyzed as follows: ； ； in, The contribution of the seasonal dimension to this misjudgment. The contribution of the load factor to this misjudgment. , This refers to the deviation of the target transformer in the current seasonal context during this misjudgment. This represents the average deviation from the historical normal state under this seasonal dimension. This refers to the deviation of the target transformer from its current load factor in this misjudgment. This represents the average deviation from historical normal conditions under this load factor dimension. Indicates the first Deviation of the target transformer in each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, That is, the deviation in the load factor dimension. Indicates the first The average deviation of the historical normal state under each dimension, when hour, That is, the deviation under the seasonal dimension; when hour, ; The total number of dimensions participating in the contribution analysis. That is, the seasonal dimension and the load factor dimension.

Images