Load data-based transformer area latent fault identification method and system

Abstract

The invention provides a transformer area latent fault identification method and system based on load data, and the method comprises the steps: employing a fault detection model, and based on the steady-state voltage difference, voltage instantaneous drop amplitude and load current in the load data of a transformer area, judging a suspected fault event in the transformer area; based on the load data of each suspected fault event, performing over-dead-zone voltage-current inverse correlation analysis and instantaneous down-frequency analysis on each suspected fault event in combination with a dynamic time warping algorithm, and determining a plurality of high-risk users corresponding to the high-risk event from the over-dead-zone voltage-current inverse correlation analysis and the instantaneous down-frequency analysis; according to the topological data of the transformer area, performing group anomaly traceability on high-risk users by using a graph neural network to determine a public fault point; resistance constant analysis is carried out on non-group high-risk users, user fault points are obtained, transformer area hidden dangers which are difficult to identify through traditional manual inspection can be effectively found, high-risk events can be accurately identified, public fault points can be deduced based on topology in combination with a graph neural network, and root cause analysis is achieved.

Description

Technical Field

[0001] This invention relates to the field of transformer substation fault identification technology, and specifically to a method and system for identifying latent faults in transformer substations based on load data. Background Technology

[0002] In power distribution network management, a distribution area (or transformer substation) refers to the entire area supplied by a single distribution transformer and is the basic unit of operation and maintenance. The distribution area extends from the distribution transformer to all users. The core power supply facilities from the distribution transformer to the user's low-voltage meter constitute the main part of the distribution area. This low-voltage link from the distribution transformer to the low-voltage meter is the capillary of the power supply chain. After long-distance transmission and voltage reduction, electricity is finally converted to 220V / 380V low-voltage electricity by the distribution transformer before being delivered to households. This "last mile" network facing the end user is the largest and most complex; a fault at any point can directly cause a power outage for the user.

[0003] Unlike overhead, professionally maintained medium- and high-voltage lines, low-voltage lines are often closer to the ground and residential areas, operating in a more complex and harsher environment, and are more susceptible to external influences. Among these, pre-meter faults are a significant factor affecting low-voltage reliability. For example, in 2024, a county power supply company received 716 repair requests due to power equipment failures, of which 606 (84%) were caused by pre-meter faults. "Pre-meter" refers to all stages from the distribution transformer, through the low-voltage main line and service connection line, to the user's meter. Low-voltage pre-meter faults in a distribution area generally have a certain latency period. How to detect and eliminate them early, effectively reduce the number of repair requests, and improve low-voltage reliability has become an urgent problem to be solved.

[0004] The existing fault investigation mode for transformer substations is periodic inspection. However, due to the harsh operating environment of low-voltage equipment, complex electricity consumption, surge in electricity consumption during peak summer seasons, and a large user base, manual inspections are difficult to cover all aspects. Furthermore, hidden hazards such as oxidation of terminals inside the meter box seals and concealed damage to cables are difficult to identify visually during inspections. As a result, low-voltage faults still occur frequently in transformer substations. These problems seriously restrict the overall improvement of the safe operation level of the distribution network. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention proposes a method for identifying latent faults in transformer substations based on load data, comprising: Using a fault detection model, based on the steady-state voltage difference, instantaneous voltage drop, and load current in the load data of the distribution area, multiple suspected fault events in the distribution area are judged in real time. Based on the load data of each suspected fault event, and combined with the dynamic time warping algorithm, the dead zone voltage-current inverse correlation analysis and transient frequency analysis are performed on each suspected fault event to identify multiple high-risk users corresponding to high-risk events. Based on the load data and topology data of the transformer area, a graph neural network is used to trace the source of group anomalies of multiple high-risk users based on spatiotemporal correlation and topology propagation to determine the common fault point of the transformer area; resistance constant analysis is performed on each of the non-group high-risk users to obtain the user fault point of the transformer area.

[0006] Optionally, based on the load data of each suspected fault event, and combined with the dynamic time warping algorithm, dead-zone voltage-current inverse correlation analysis and transient frequency analysis are performed on each suspected fault event to identify multiple high-risk users corresponding to high-risk events, including: Based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; The inverse correlation between voltage and current curves in the dead zone within the time window is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value; and the theoretical abnormal resistance is calculated using the voltage change and load current during the instantaneous voltage drop. The frequency of voltage drop, the magnitude of voltage drop, the duration of low voltage, the inverse correlation value, the theoretical abnormal resistance, and the characteristics of power outage and overheating risks are integrated into a multi-dimensional feature, which is then input into a risk quantification model to obtain a risk value. Based on the risk value, it is determined whether the suspected fault event is a high-risk event, and the user corresponding to the high-risk event is recorded as a high-risk user.

[0007] Optionally, the power outage and overheating risk characteristics are obtained in the following manner: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the association result of whether it is related to the power outage event; Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. Combining the correlation results and melting risk values, the characteristics of power outage and overheating risks are obtained.

[0008] Optionally, the risk quantification model is trained in the following manner: Obtain high-risk and low-risk event data from historical event data, and label them as high-risk and low-risk to obtain the training dataset; Based on the training dataset, the instantaneous drop frequency, instantaneous drop amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics of each historical event data are used as input features of the model, and high-risk labels and low-risk labels are used as prediction targets to train the random forest model, thereby obtaining a trained risk quantification model. The output of the random forest model includes the prediction target and the prediction probability, where the prediction probability is a risk value.

[0009] Optionally, the fault detection model, based on the steady-state voltage difference, instantaneous voltage drop, and load current in the load data of the distribution area, is used to determine multiple suspected fault events in the distribution area in real time, including: The fault detection model is used to determine whether the instantaneous voltage drop of each user's electricity meter in the load data of the distribution area exceeds the voltage drop threshold. For multiple user meters that exceed the drop amplitude threshold, determine whether the load current of each user meter is greater than the preset current. If it is greater, further determine whether the steady-state voltage difference between each user meter and the total meter of the distribution area exceeds the voltage difference threshold. If the differential pressure threshold is exceeded, the theoretical power is calculated based on the voltage and load current of each user's electricity meter at this time, and it is determined whether the theoretical power is lower than the power threshold. If it is lower, a snapshot of the complete electricity consumption data of the user's electricity meter at this time is taken as a suspected fault event.

[0010] Optionally, the step of using a graph neural network to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users, and determining the common fault point of the transformer area, based on the load data and topology data of the transformer area, includes: Based on the topology data of the distribution area, the transformers, branch boxes and user meters in the distribution area are abstracted as nodes, the electrical connection relationship between each node is abstracted as an edge, and the load data of each node is used as a node feature to construct the graph structure of the distribution area. The instantaneous drop amplitude and voltage instantaneous drop time in the load data of the high-risk users are injected as abnormal features into the corresponding nodes in the graph structure; By using a graph neural network and a message passing mechanism, the abnormal features of downstream nodes in the graph structure are propagated along the edges to upstream nodes to perform group anomaly tracing, obtain the final abnormal upstream node, and record the propagation path. The abnormal upstream node is designated as the common fault point of the transformer area, and all nodes on the propagation path are recorded as users experiencing a group fault.

[0011] Optionally, the step of performing constant resistance analysis on each of the non-group high-risk users to obtain the user fault points in the transformer area includes: Based on each of the aforementioned high-risk users who are not part of a group: Using Ohm's law, based on the voltage drop amplitude and load current during each voltage instantaneous drop for the high-risk user, the resistance at each voltage instantaneous drop is calculated, resulting in multiple fault resistors; If the values ​​of multiple fault resistors fluctuate within a preset range, then the user fault point of the high-risk user is determined to be a user-end fault.

[0012] Based on the same inventive concept, this invention proposes a transformer substation latent fault identification system based on load data, comprising: The preliminary fault judgment module is used to use a fault detection model to judge multiple suspected fault events in the transformer area in real time based on the steady-state voltage difference, instantaneous voltage drop amplitude and load current in the load data of the transformer area. The high-risk user screening module is used to perform dead-zone voltage-current inverse correlation analysis and transient frequency analysis on each suspected fault event based on the load data of each suspected fault event, combined with the dynamic time warping algorithm, to identify multiple high-risk users corresponding to the high-risk events. The fault point analysis module is used to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users using graph neural networks, based on the load data and topology data of the transformer area, to determine the common fault point of the transformer area; and to perform constant resistance analysis on each of the non-group high-risk users to obtain the user fault point of the transformer area.

[0013] Optionally, the high-risk user screening module is specifically used for: Based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; The inverse correlation between voltage and current curves in the dead zone within the time window is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value; and the theoretical abnormal resistance is calculated using the voltage change and load current during the instantaneous voltage drop. The frequency of voltage drop, the magnitude of voltage drop, the duration of low voltage, the inverse correlation value, the theoretical abnormal resistance, and the characteristics of power outage and overheating risks are integrated into a multi-dimensional feature, which is then input into a risk quantification model to obtain a risk value. Based on the risk value, it is determined whether the suspected fault event is a high-risk event, and the user corresponding to the high-risk event is recorded as a high-risk user.

[0014] Optionally, the high-risk user screening module is specifically used for: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the association result of whether it is related to the power outage event; Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. Combining the correlation results and melting risk values, the characteristics of power outage and overheating risks are obtained.

[0015] Optionally, the high-risk user screening module is further used for: Obtain high-risk and low-risk event data from historical event data, and label them as high-risk and low-risk to obtain the training dataset; Based on the training dataset, the instantaneous drop frequency, instantaneous drop amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics of each historical event data are used as input features of the model, and high-risk labels and low-risk labels are used as prediction targets to train the random forest model, thereby obtaining a trained risk quantification model. The output of the random forest model includes the prediction target and the prediction probability, where the prediction probability is a risk value.

[0016] Optionally, the preliminary fault diagnosis module is specifically used for: The fault detection model is used to determine whether the instantaneous voltage drop of each user's electricity meter in the load data of the distribution area exceeds the voltage drop threshold. For multiple user meters that exceed the drop amplitude threshold, determine whether the load current of each user meter is greater than the preset current. If it is greater, further determine whether the steady-state voltage difference between each user meter and the total meter of the distribution area exceeds the voltage difference threshold. If the differential pressure threshold is exceeded, the theoretical power is calculated based on the voltage and load current of each user's electricity meter at this time, and it is determined whether the theoretical power is lower than the power threshold. If it is lower, a snapshot of the complete electricity consumption data of the user's electricity meter at this time is taken as a suspected fault event.

[0017] Optionally, the fault point analysis module is specifically used for: Based on the topology data of the distribution area, the transformers, branch boxes and user meters in the distribution area are abstracted as nodes, the electrical connection relationship between each node is abstracted as an edge, and the load data of each node is used as a node feature to construct the graph structure of the distribution area. The instantaneous drop amplitude and voltage instantaneous drop time in the load data of the high-risk users are injected as abnormal features into the corresponding nodes in the graph structure; By using a graph neural network and a message passing mechanism, the abnormal features of downstream nodes in the graph structure are propagated along the edges to upstream nodes to perform group anomaly tracing, obtain the final abnormal upstream node, and record the propagation path. The abnormal upstream node is designated as the common fault point of the transformer area, and all nodes on the propagation path are recorded as users experiencing a group fault.

[0018] Optionally, the fault point analysis module is specifically used for: Based on each of the aforementioned high-risk users who are not part of a group: Using Ohm's law, based on the voltage drop amplitude and load current during each voltage instantaneous drop for the high-risk user, the resistance at each voltage instantaneous drop is calculated, resulting in multiple fault resistors; If the values ​​of multiple fault resistors fluctuate within a preset range, then the user fault point of the high-risk user is determined to be a user-end fault.

[0019] In another aspect, this application also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a method for identifying latent faults in transformer substations based on load data, as described above, is implemented.

[0020] In another aspect, this application also provides a computer-readable storage medium having an executable program stored thereon, which, when executed, implements the method for identifying latent faults in transformer substations based on load data as described above.

[0021] Compared with the closest existing technology, the present invention has the following beneficial effects: This invention provides a method and system for identifying latent faults in transformer substations based on load data, comprising: using a fault detection model, judging multiple suspected fault events in the substation in real time based on the steady-state voltage difference, instantaneous voltage drop amplitude, and load current in the load data of the substation; based on the load data of each suspected fault event, performing dead-zone voltage-current inverse correlation analysis and instantaneous drop frequency analysis on each suspected fault event using a dynamic time warping algorithm, thereby identifying multiple high-risk users corresponding to high-risk events; based on the load data and topology data of the substation, using a graph neural network to perform group anomaly tracing based on spatiotemporal correlation and topology propagation on the multiple high-risk users, determining the common fault point of the substation; and performing constant resistance analysis on each of the non-group high-risk users to obtain the user fault point of the substation. Since faults in power grid operation manifest as a rise in current accompanied by a significant drop in voltage, this solution uses real-time monitoring of voltage dips and load current in transformer substations to identify suspected fault events. This effectively uncovers hidden hazards that are difficult to identify through traditional manual inspections, enabling early diagnosis of latent faults. It no longer relies on late-stage phenomena such as heat and smoke observed by the naked eye, but instead identifies the abnormally high impedance of the fault through electrical characteristics, thus significantly advancing the fault detection time and ensuring the safe operation of the power grid. Furthermore, this solution uses a dynamic time warping algorithm to solve the problem of misjudgment caused by asynchronous transmission of voltage and current data, accurately identifying high-risk events caused by load surges. Utilizing graph neural networks, it can reverse-engineer common fault points based on transformer substation topology, enabling root cause analysis. This allows for the location and handling of safety hazards before a single common fault point triggers a large-scale power outage, reducing the fault incidence rate at its source. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a method for identifying latent faults in transformer substations based on load data, provided by this invention. Figure 1 ; Figure 2 A flowchart illustrating the method for identifying latent faults in transformer substations based on load data provided by the present invention (Figure 2). Figure 3 A schematic diagram of a latent fault identification system for transformer substations based on load data provided by the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided by the present invention. Detailed Implementation

[0023] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0024] Example 1 This invention provides a method for identifying latent faults in transformer substations based on load data, such as... Figure 1As shown, it includes: S1. Using a fault detection model, based on the steady-state voltage difference, instantaneous voltage drop amplitude, and load current in the load data of the distribution area, multiple suspected fault events in the distribution area are judged in real time. S2. Based on the load data of each suspected fault event, and combined with the dynamic time warping algorithm, perform dead zone voltage-current inverse correlation analysis and transient frequency analysis on each suspected fault event to identify multiple high-risk users corresponding to high-risk events. S3. Based on the load data and topology data of the transformer area, a graph neural network is used to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users to determine the common fault point of the transformer area; resistance constant analysis is performed on each of the non-group high-risk users to obtain the user fault point of the transformer area.

[0025] Combination Figure 2 This plan will be explained in detail.

[0026] In step S1, the fault detection model is used to determine the instantaneous voltage drop of each user's meter in the load data of the distribution area (i.e., Figure 2 Whether the voltage drop at the measurement points before and after the test exceeds the voltage drop threshold; For multiple user meters that exceed the drop amplitude threshold, the fault detection model is used to determine whether the load current of each user meter is greater than the preset current. If it is greater, it is further determined whether the steady-state voltage difference between each user meter and the total meter of the distribution area exceeds the voltage difference threshold. If the differential pressure threshold is exceeded, the theoretical power is calculated based on the voltage and load current of each user's electricity meter at this time, and it is determined whether the theoretical power is lower than the power threshold. If it is lower, a snapshot of the complete electricity consumption data of the user's electricity meter at this time is taken as a suspected fault event.

[0027] Specifically, the basic unit of real-time detection is each user's electricity meter, and detection is carried out on the data stream of each user's electricity meter. Using a fault detection model, the load data stream of each user's electricity meter is monitored in real time. When the voltage value at a certain moment T0 is detected to have dropped more than a threshold compared to the previous moment, moment T0 is marked as the starting point of a potential "voltage sag event" and proceeds to subsequent judgment. The load current recorded by the user's electricity meter at the moment of the sag T0 is obtained, and it is determined whether it is greater than a preset current (set to 0.1A in this embodiment). This ensures that the user's load is indeed running when the event occurs. Based on this, the voltage value of the user at moment T0 is compared with the voltage value of the main meter of the distribution area. If the difference between the two (voltage difference) exceeds the threshold, it indicates that the voltage drop mainly occurs on the user's branch. For user meters that exceed the voltage difference threshold, the theoretical power is calculated based on the user's current voltage and load current at this time, and the average power of the user's meter during normal historical periods is obtained as a benchmark. If the theoretical power at this time is lower than 80% of the average power, it indicates that at the same time as the voltage drop, the user's equipment may not be able to reach normal power due to low voltage, which is a typical feature of a fault event. In this case, a snapshot of the complete electricity consumption data of the user's meter at this time is taken as a suspected fault event. The suspected fault event includes the event subject identifier, event timestamp, event load characteristic data, and event conclusion.

[0028] In the above steps, once the moment of instantaneous voltage drop is identified, the subsequent judgments are all based on the data at this critical moment, comprehensively analyzing whether a suspected fault has occurred, and obtaining more accurate results.

[0029] In step S2, based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; Based on the voltage and current of the user's electricity meter within the time window, the inverse correlation of the voltage and current curves in the dead zone is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value. The theoretical abnormal resistance is calculated using the voltage change and load current during the voltage dip of the suspected fault event. The instantaneous descent frequency, instantaneous descent amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics are integrated into multi-dimensional features and input into the risk quantification model to conduct risk assessment and obtain risk values. Based on the risk value, determine whether the suspected fault event is a high-risk event, and record the user corresponding to the high-risk event as a high-risk user.

[0030] The power outage and overheating risk characteristics are obtained in the following manner: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the correlation result of whether it is related to the power outage event. The output of the correlation result is a Boolean type, with a result of 0 (no correlation) or 1 (correlation). The power outage correlation result analysis is to analyze whether the voltage drop is likely not a problem on the user side, but caused by a power outage on the grid side. Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. By combining the correlation results and the melting risk value, the characteristics of power outage and overheating risks are obtained. These characteristics are analytical features extracted based on expert knowledge.

[0031] The risk quantification model is trained in the following manner: Obtain high-risk and low-risk event data from historical event data, and label them as high-risk and low-risk to obtain the training dataset; Based on the training dataset, the instantaneous drop frequency, instantaneous drop amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics of each historical event data are used as input features of the model, and high-risk labels and low-risk labels are used as prediction targets to train the random forest model, thereby obtaining a trained risk quantification model. The output of the random forest model includes the prediction target and the prediction probability, where the prediction probability is a risk value.

[0032] When the above-mentioned frequency of transient voltage drop, amplitude of transient voltage drop, duration of low voltage, inverse correlation value and theoretical abnormal resistance are fused into a multi-feature calculation and input into the risk quantification model for risk assessment, the model will determine whether the suspected fault event belongs to the high-risk or low-risk category. At the same time, the random forest model will count the voting results of all decision trees in its internal system. If there are 100 trees, and 85 of them consider the event to be "high-risk", then the model will output the probability of "high risk" as 85%, that is, the risk value is 85%.

[0033] This approach defines high risk as the positive class during model training. Therefore, the model's classification decision will be high risk (because 0.85 > 0.5), not low risk. If the model outputs low risk, it actually means the calculated probability of high risk is very low, for example, P(high risk) = 0.15. Then, the probability of belonging to low risk is P(low risk) = 1 - P(high risk) = 0.85. In this case, the risk value should be P(high risk) = 0.15, not P(low risk) = 0.85.

[0034] This step further quantifies the risks of suspected fault events initially screened by user meters by integrating multi-dimensional features, and builds a comprehensive assessment system. This overcomes the limitations of a single criterion, enables accurate identification of faults and quantification of fault severity, helps maintenance personnel to clearly warn of high-risk hazards, improves maintenance security, and reduces the misjudgment rate.

[0035] In step S3, S3-1, based on the load data and topology data of the transformer area, a graph neural network is used to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users to determine the common fault point of the transformer area.

[0036] Based on the topology data of the distribution area, the transformers, branch boxes and user meters in the distribution area are abstracted as nodes, the electrical connection relationship between each node is abstracted as an edge, and the load data of each node is used as a node feature to construct the graph structure of the distribution area. The instantaneous drop amplitude and voltage instantaneous drop time in the load data of the high-risk users are injected as abnormal features into the corresponding nodes in the graph structure; By using a graph neural network and a message passing mechanism, the abnormal features of downstream nodes in the graph structure are propagated along the edges to upstream nodes to perform group anomaly tracing, obtain the final abnormal upstream node, and record the propagation path. The abnormal upstream node is designated as the common fault point of the transformer area, and all nodes on the propagation path are recorded as users experiencing a group fault.

[0037] Graph neural networks can automatically and accurately locate the upstream common fault point that causes group problems from numerous scattered user anomalies, and focus the maintenance target from a group of users to a single device. This realizes the transformation from passively responding to a large number of user repair reports to proactively eliminating a few common fault sources, thus eliminating noise interference and improving operation and maintenance efficiency and accuracy.

[0038] S3-2. Perform constant resistance analysis on each of the non-group high-risk users to obtain the user fault points in the transformer area.

[0039] The reason for using constant resistance in the analysis is that the resistance of a fixed fault point (such as a loose connector at the front end of a user's electricity meter) is relatively constant in the short term.

[0040] Based on each of the aforementioned high-risk users who are not part of a group: Using Ohm's law, based on the voltage drop amplitude and load current during each voltage instantaneous drop for the high-risk user, the resistance at each voltage instantaneous drop is calculated, resulting in multiple fault resistors; If the values ​​of multiple fault resistors fluctuate within a preset range, then the user fault point of the high-risk user is determined to be a user-end fault.

[0041] In other words, if the calculated series of fault resistance values ​​fluctuate within a narrow, preset range (i.e., remain essentially constant), this constitutes strong evidence that these voltage drops are all caused by the same physical fault point (i.e., an abnormal resistor with a constant resistance). Using this method, the fault can be directly located on the user's dedicated power supply line, such as the incoming line connector in the user's meter box, or a section of the line from the branch box to the user's meter, thus identifying the user's fault point.

[0042] Optionally, the inverse correlation between current and voltage can be added to the user-end fault diagnosis.

[0043] In distribution network operation and maintenance, a user-end fault specifically refers to a fault that occurs before a user's electricity meter but belongs to the user's dedicated power supply facility.

[0044] This solution, after identifying user-end faults and common fault points, also includes: pushing potential hazard information to local power supply station maintenance personnel via SMS platform; after on-site inspection and repair, filling in the fault elimination record; and the system automatically verifying whether the fault elimination was successful, completing the push and detection closed loop. Figure 2 The unified access authentication refers to the system using a unified and centralized access control module to verify and confirm whether the current maintenance personnel are qualified to receive the information and perform the troubleshooting task when pushing fault and potential hazard information to the local power supply station's maintenance personnel.

[0045] Example 2 Based on the same inventive concept, this invention also provides a transformer substation latent fault identification system based on load data, such as... Figure 3 As shown, it includes: The preliminary fault judgment module is used to use a fault detection model to judge multiple suspected fault events in the transformer area in real time based on the steady-state voltage difference, instantaneous voltage drop amplitude and load current in the load data of the transformer area. The high-risk user screening module is used to perform dead-zone voltage-current inverse correlation analysis and transient frequency analysis on each suspected fault event based on the load data of each suspected fault event, combined with the dynamic time warping algorithm, to identify multiple high-risk users corresponding to the high-risk events. The fault point analysis module is used to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users using graph neural networks, based on the load data and topology data of the transformer area, to determine the common fault point of the transformer area; and to perform constant resistance analysis on each of the non-group high-risk users to obtain the user fault point of the transformer area.

[0046] In one possible implementation, the aforementioned high-risk user screening module is specifically used for: Based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; The inverse correlation between voltage and current curves in the dead zone within the time window is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value; and the theoretical abnormal resistance is calculated using the voltage change and load current during the instantaneous voltage drop. The frequency of voltage drop, the magnitude of voltage drop, the duration of low voltage, the inverse correlation value, the theoretical abnormal resistance, and the characteristics of power outage and overheating risks are integrated into a multi-dimensional feature, which is then input into a risk quantification model to obtain a risk value. Based on the risk value, it is determined whether the suspected fault event is a high-risk event, and the user corresponding to the high-risk event is recorded as a high-risk user.

[0047] In one possible implementation, the aforementioned high-risk user screening module is specifically used for: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the association result of whether it is related to the power outage event; Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. Combining the correlation results and melting risk values, the characteristics of power outage and overheating risks are obtained.

[0048] In one possible implementation, the high-risk user screening module described above is also used for: Obtain high-risk and low-risk event data from historical event data, and label them as high-risk and low-risk to obtain the training dataset; Based on the training dataset, the instantaneous drop frequency, instantaneous drop amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics of each historical event data are used as input features of the model, and high-risk labels and low-risk labels are used as prediction targets to train the random forest model, thereby obtaining a trained risk quantification model. The output of the random forest model includes the prediction target and the prediction probability, where the prediction probability is a risk value.

[0049] In one possible implementation, the aforementioned preliminary fault diagnosis module is specifically used for: The fault detection model is used to determine whether the instantaneous voltage drop of each user's electricity meter in the load data of the distribution area exceeds the voltage drop threshold. For multiple user meters that exceed the drop amplitude threshold, determine whether the load current of each user meter is greater than the preset current. If it is greater, further determine whether the steady-state voltage difference between each user meter and the total meter of the distribution area exceeds the voltage difference threshold. If the differential pressure threshold is exceeded, the theoretical power is calculated based on the voltage and load current of each user's electricity meter at this time, and it is determined whether the theoretical power is lower than the power threshold. If it is lower, a snapshot of the complete electricity consumption data of the user's electricity meter at this time is taken as a suspected fault event.

[0050] In one possible implementation, the aforementioned fault point analysis module is specifically used for: Based on the topology data of the distribution area, the transformers, branch boxes and user meters in the distribution area are abstracted as nodes, the electrical connection relationship between each node is abstracted as an edge, and the load data of each node is used as a node feature to construct the graph structure of the distribution area. The instantaneous drop amplitude and voltage instantaneous drop time in the load data of the high-risk users are injected as abnormal features into the corresponding nodes in the graph structure; By using a graph neural network and a message passing mechanism, the abnormal features of downstream nodes in the graph structure are propagated along the edges to upstream nodes to perform group anomaly tracing, obtain the final abnormal upstream node, and record the propagation path. The abnormal upstream node is designated as the common fault point of the transformer area, and all nodes on the propagation path are recorded as users experiencing a group fault.

[0051] In one possible implementation, the aforementioned fault point analysis module is specifically used for: Based on each of the aforementioned high-risk users who are not part of a group: Using Ohm's law, based on the voltage drop amplitude and load current during each voltage instantaneous drop for the high-risk user, the resistance at each voltage instantaneous drop is calculated, resulting in multiple fault resistors; If the values ​​of multiple fault resistors fluctuate within a preset range, then the user fault point of the high-risk user is determined to be a user-end fault.

[0052] Example 3 like Figure 4As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0053] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to realize the corresponding method flow or corresponding function, so as to realize the steps of the method for identifying latent faults in transformer areas based on load data in the above embodiments.

[0054] Example 4 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the method for identifying latent faults in transformer substations based on load data in the above embodiments.

[0055] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0057] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0058] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the claims pending approval.

Claims

1. A method for identifying latent faults in transformer substations based on load data, characterized in that, include: Using a fault detection model, based on the steady-state voltage difference, instantaneous voltage drop, and load current in the load data of the distribution area, multiple suspected fault events in the distribution area are judged in real time. Based on the load data of each suspected fault event, and combined with the dynamic time warping algorithm, the dead zone voltage-current inverse correlation analysis and transient frequency analysis are performed on each suspected fault event to identify multiple high-risk users corresponding to high-risk events. Based on the load data and topology data of the transformer area, a graph neural network is used to trace the source of group anomalies of multiple high-risk users based on spatiotemporal correlation and topology propagation, and to determine the common fault point of the transformer area. By performing constant resistance analysis on each of the non-group high-risk users, the user fault points in the transformer area can be obtained.

2. The method as described in claim 1, characterized in that, Based on the load data of each suspected fault event, and combined with the dynamic time warping algorithm, dead-zone voltage-current inverse correlation analysis and transient frequency analysis are performed on each suspected fault event to identify multiple high-risk users corresponding to high-risk events, including: Based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; The inverse correlation between voltage and current curves in the dead zone within the time window is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value; and the theoretical abnormal resistance is calculated using the voltage change and load current during the instantaneous voltage drop. The frequency of voltage drop, the magnitude of voltage drop, the duration of low voltage, the inverse correlation value, the theoretical abnormal resistance, and the characteristics of power outage and overheating risks are integrated into a multi-dimensional feature, which is then input into a risk quantification model to obtain a risk value. Based on the risk value, it is determined whether the suspected fault event is a high-risk event, and the user corresponding to the high-risk event is recorded as a high-risk user.

3. The method as described in claim 2, characterized in that, The characteristics of power outage and overheating risks are obtained in the following manner: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the association result of whether it is related to the power outage event; Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. Combining the correlation results and melting risk values, the characteristics of power outage and overheating risks are obtained.

4. The method as described in claim 2, characterized in that, The risk quantification model is trained in the following manner: Obtain high-risk and low-risk event data from historical event data, and label them as high-risk and low-risk to obtain the training dataset; Based on the training dataset, the instantaneous drop frequency, instantaneous drop amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics of each historical event data are used as input features of the model, and high-risk labels and low-risk labels are used as prediction targets to train the random forest model, thereby obtaining a trained risk quantification model. The output of the random forest model includes the prediction target and the prediction probability, where the prediction probability is a risk value.

5. The method as described in claim 1, characterized in that, The fault detection model, based on the steady-state voltage difference, instantaneous voltage drop, and load current in the transformer substation load data, is used to determine multiple suspected fault events in the transformer substation in real time, including: The fault detection model is used to determine whether the instantaneous voltage drop of each user's electricity meter in the load data of the distribution area exceeds the voltage drop threshold. For multiple user meters that exceed the drop amplitude threshold, determine whether the load current of each user meter is greater than the preset current. If it is greater, further determine whether the steady-state voltage difference between each user meter and the total meter of the distribution area exceeds the voltage difference threshold. If the differential pressure threshold is exceeded, the theoretical power is calculated based on the voltage and load current of each user's electricity meter at this time, and it is determined whether the theoretical power is lower than the power threshold. If it is lower, a snapshot of the complete electricity consumption data of the user's electricity meter at this time is taken as a suspected fault event.

6. The method as described in claim 1, characterized in that, The step of using a graph neural network to trace the source of group anomalies among multiple high-risk users based on spatiotemporal correlation and topology propagation, according to the load data and topology data of the transformer area, to determine the common fault point of the transformer area includes: Based on the topology data of the distribution area, the transformers, branch boxes and user meters in the distribution area are abstracted as nodes, the electrical connection relationship between each node is abstracted as an edge, and the load data of each node is used as a node feature to construct the graph structure of the distribution area. The instantaneous drop amplitude and voltage instantaneous drop time in the load data of the high-risk users are injected as abnormal features into the corresponding nodes in the graph structure; By using a graph neural network and a message passing mechanism, the abnormal features of downstream nodes in the graph structure are propagated along the edges to upstream nodes to perform group anomaly tracing, obtain the final abnormal upstream node, and record the propagation path. The abnormal upstream node is designated as the common fault point of the transformer area, and all nodes on the propagation path are recorded as users experiencing a group fault.

7. The method as described in claim 1, characterized in that, The method of performing constant resistance analysis on each of the non-group high-risk users to obtain the user fault points in the transformer area includes: Based on each of the aforementioned high-risk users who are not part of a group: Using Ohm's law, based on the voltage drop amplitude and load current during each voltage instantaneous drop for the high-risk user, the resistance at each voltage instantaneous drop is calculated, resulting in multiple fault resistors; If the values ​​of multiple fault resistors fluctuate within a preset range, then the user fault point of the high-risk user is determined to be a user-end fault.

8. A latent fault identification system for transformer substations based on load data, characterized in that, include: The preliminary fault judgment module is used to use a fault detection model to judge multiple suspected fault events in the transformer area in real time based on the steady-state voltage difference, instantaneous voltage drop amplitude and load current in the load data of the transformer area. The high-risk user screening module is used to perform dead-zone voltage-current inverse correlation analysis and transient frequency analysis on each suspected fault event based on the load data of each suspected fault event, combined with the dynamic time warping algorithm, to identify multiple high-risk users corresponding to the high-risk events. The fault point analysis module is used to perform group anomaly tracing based on spatiotemporal correlation and topology propagation for multiple high-risk users using graph neural networks, based on the load data and topology data of the transformer area, to determine the common fault point of the transformer area. By performing constant resistance analysis on each of the non-group high-risk users, the user fault points in the transformer area can be obtained.

9. The system as described in claim 8, characterized in that, The high-risk user screening module is specifically used for: Based on each suspected fault event: Extract the instantaneous descent frequency, instantaneous descent amplitude, and low voltage duration from the load data of the user's electricity meter corresponding to the suspected fault event within a preset time window; The inverse correlation between voltage and current curves in the dead zone within the time window is analyzed using the dynamic time warping algorithm to obtain the inverse correlation value; and the theoretical abnormal resistance is calculated using the voltage change and load current during the instantaneous voltage drop. The instantaneous descent frequency, instantaneous descent amplitude, low voltage duration, inverse correlation value, theoretical abnormal resistance, and power outage and overheating risk characteristics are integrated into multi-dimensional features and input into the risk quantification model to conduct risk assessment and obtain risk values. Based on the risk value, determine whether the suspected fault event is a high-risk event, and record the user corresponding to the high-risk event as a high-risk user.

10. The system as described in claim 8, characterized in that, The high-risk user screening module is specifically used for: The voltage drop time of the suspected fault event is compared with the timestamp of the power outage event in the transformer area to obtain the association result of whether it is related to the power outage event; Based on the theoretical abnormal resistance and the load current during the instantaneous voltage drop, the abnormal heat generation of the suspected fault event is calculated by combining Joule's law. The abnormal heat generation is then compared with the melting point temperature threshold of the user's meter connector material to obtain the melting risk value of the user's meter connector. Combining the correlation results and melting risk values, the characteristics of power outage and overheating risks are obtained.

Images