A method, apparatus, and equipment for power equipment anomaly identification based on spatiotemporal convolutional networks.

Abstract

This invention provides a method, apparatus, and device for power equipment anomaly identification based on spatiotemporal convolutional networks, belonging to the field of anomaly identification technology. The method includes: obtaining the interval adjustment amount corresponding to each operational data item of each power device and determining a baseline normal interval based on the topology of the target power grid and operational sequence data within a preset time window; comparing the real-time operational data in the operational sequence data of each power device with the baseline normal interval to obtain the abnormal operational data of each power device, and determining the anomaly propagation path based on the topology of each power device; adjusting the abnormal operational data of each power device along the anomaly propagation path based on the abnormal operational data of all power devices along the anomaly propagation path, and determining the anomaly information of each power device based on the adjustment result. This invention enables anomaly information to effectively adapt to the dynamic operating conditions of the power system, effectively improving the accuracy of power equipment anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of anomaly recognition technology, and in particular to a method, apparatus, and device for anomaly recognition of power equipment based on spatiotemporal convolutional networks. Background Technology

[0002] The large-scale time-series data generated during the operation of power equipment in a power system is a digital mapping of the equipment's full-condition operating status. It encompasses multi-dimensional information such as equipment measurement data, operating parameters, and condition monitoring indicators, containing core characteristics of equipment health status, operational trends, and potential faults. This data is crucial for ensuring the safe and stable operation of the power system. Accurately extracting valuable information from time-series data can proactively detect equipment operational risks and predict fault trends, providing data support for grid dispatching decisions and lean equipment maintenance. This has irreplaceable value in improving the overall reliability of the power system.

[0003] In existing technologies, static threshold detection methods are typically used to locate abnormal features in time series data. This method requires combining the rated operating parameters of power equipment and time series data samples under historical normal operating conditions to define the normal value range and static threshold range of each monitoring indicator. Then, the real-time collected time series data is compared and analyzed point by point, and data points that exceed the preset threshold range are judged as abnormal data. Anomaly identification is achieved based on the correlation between the monitoring point and the equipment corresponding to the data.

[0004] However, existing technologies rely on fixed numerical ranges, which cannot adapt to the dynamic operating conditions of power system load fluctuations and environmental changes, and are prone to misjudgment. Summary of the Invention

[0005] This invention provides a method, apparatus, and device for identifying power equipment anomalies based on spatiotemporal convolutional networks, in order to solve the problem that existing technologies rely on fixed numerical ranges, which cannot adapt to the dynamic operating conditions of power system load fluctuations and environmental changes, and are prone to misjudgment.

[0006] In a first aspect, embodiments of the present invention provide a method for identifying power equipment anomalies based on spatiotemporal convolutional networks, including: The topology of the target power grid and the runtime sequence data within a preset time window are input into a pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each operational data of each power device, and the baseline normal interval is determined based on the interval adjustment amount. The real-time operating data in the runtime sequence data of each power device is compared with the corresponding baseline normal range to obtain the abnormal operating data of each power device. Based on the topology and abnormal operating data of each power device, at least one abnormal propagation path of the target power grid is determined. Based on the abnormal operation data of all power equipment along each abnormal propagation path, the abnormal operation data of each power equipment along the abnormal propagation path is adjusted, and the abnormal information of each power equipment is determined based on the adjustment results of each power equipment.

[0007] Secondly, embodiments of the present invention provide a power equipment anomaly identification device based on a spatiotemporal convolutional network, comprising: The determination module is used to input the topology of the target power grid and the runtime sequence data within a preset time window into a pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each operational data of each power device, and determine the benchmark normal interval based on the interval adjustment amount. The comparison module is used to compare the real-time operating data in the runtime sequence data of each power device with the corresponding baseline normal range to obtain the abnormal operating data of each power device, and to determine at least one abnormal propagation path of the target power grid based on the topology and abnormal operating data of each power device. The adjustment module is used to adjust the abnormal operation data of each power device on each abnormal propagation path based on the abnormal operation data of all power devices on each abnormal propagation path, and determine the abnormal information of each power device based on the adjustment results of each power device.

[0008] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0009] In this embodiment of the invention, the topology of the target power grid and the runtime sequence data within a preset time window are input into a pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount. The baseline normal range of the operating data of each power device can be dynamically determined based on the interval adjustment amount. Then, by comparing and identifying abnormal operating data with real-time operating data and determining the abnormal propagation path based on the topology, the abnormal information of each power device can be accurately determined by adjusting the abnormal operating data on the abnormal propagation path. This can effectively adapt to the dynamic operating conditions brought about by power system load fluctuations and environmental changes, effectively improve the accuracy and pertinence of power device abnormal identification, and provide reliable support for the safe and stable operation of the power grid. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the implementation of the power equipment anomaly identification method based on spatiotemporal convolutional networks provided in this embodiment of the invention. Figure 2 This is a flowchart illustrating the implementation of step S130 of the power equipment anomaly identification method based on spatiotemporal convolutional networks provided in this embodiment of the invention. Figure 3 This is a schematic diagram of the structure of the power equipment anomaly identification device based on spatiotemporal convolutional network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0011] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0012] See Figure 1 The document illustrates a flowchart of the implementation of the power equipment anomaly identification method based on spatiotemporal convolutional networks provided in an embodiment of the present invention, detailed below: Step S110: Input the topology of the target power grid and the runtime sequence data within the preset time window into the pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each running data of each power device, and determine the benchmark normal interval based on the interval adjustment amount.

[0013] In some embodiments, the target power grid is a specific power network used for anomaly identification. The target power grid contains various electrical equipment and electrical connection lines, forming a complete power system unit. Its scope can be flexibly defined according to actual application needs. For example, the distribution network of a city's main urban area or the dedicated power supply network of an industrial park can both serve as the target power grid to be detected. All anomaly identification operations in this method revolve around the operating status of this specific power grid. Topology refers to the electrical connection relationships and spatial layout structure between various electrical equipment in the target power grid, including the connection methods, hierarchical relationships, and networking forms between electrical equipment. It is an important basis for determining the propagation path of anomalies within the power grid. For example, the series connection relationship between substations and distribution transformers, and the parallel connection structure of multiple switchgear and busbars, are specific manifestations of the power grid topology.

[0014] In some embodiments, a preset time window is a continuous time range pre-defined for collecting power equipment operation data. Its duration can be set according to the operating characteristics of the power equipment and the actual needs of anomaly detection. The operation data collected within this time range can fully reflect the changes in the equipment's operating status over a short period. For example, a one-hour time window can be set for a target power grid including high-voltage transmission equipment. Operational sequence data refers to various types of operational status-related data continuously collected in chronological order during the operation of the power equipment. Examples include real-time oil temperature data of transformers, circuit breaker opening and closing frequency data, and real-time current and voltage data of transmission lines. This type of data forms a continuous time sequence over time.

[0015] It should be noted that before being applied to power equipment anomaly identification, the spatiotemporal convolutional network has undergone parameter learning and optimization using a large amount of relevant training data. This training data includes information such as the topology of the target power grid and historical operational sequence data. The pre-trained network can accurately process the spatiotemporal data of the power grid and output effective analysis results. For example, after training using nearly a year's worth of historical operational data from a power grid, the network can be adapted to the operational data characteristics of that power grid for targeted analysis. The spatiotemporal convolutional network is a convolutional neural network model that integrates spatial feature extraction and time series analysis capabilities. It can capture the spatial connection characteristics between different power devices in the target power grid and also uncover the temporal variation patterns in operational sequence data, simultaneously processing both the topological spatial information of the power grid and the time series information of the data.

[0016] In some embodiments, the interval adjustment amount refers to a quantified value output by the spatiotemporal convolutional network based on the topology and runtime sequence data of the target power grid, used to adjust the normal range of power equipment operating data. This value changes in real time according to the dynamic operating conditions of the power grid, adapting to power grid load fluctuations, environmental changes, etc. For example, when the power grid is in a peak load period, the network will output a corresponding interval adjustment amount to reasonably correct the normal range of current data, avoiding abnormal misjudgments caused by fixed ranges. The baseline normal range is determined based on the interval adjustment amount and is a normal range of power equipment operating data that adapts to the current dynamic operating conditions of the target power grid. The baseline normal range changes dynamically with changes in power grid load, environment, and other factors, and is a direct basis for determining whether the real-time operating data of power equipment is abnormal. For example, the current baseline normal range of a certain reactor will show different reasonable ranges during low-load and high-load phases of the power grid based on the interval adjustment amount.

[0017] In one possible implementation, the training data for the pre-trained spatiotemporal convolutional network includes the topology of the target power grid, the historical runtime sequence data of the target power grid, and the historical interval adjustment amount corresponding to each operational data of each power device in the target power grid.

[0018] In some embodiments, training data refers to data used for model training, parameter learning, and optimization of spatiotemporal convolutional networks. Training the spatiotemporal convolutional network with training data enables the network to adapt to power equipment anomaly identification scenarios and achieve accurate analysis and output capabilities. For example, quantitative data related to various power grid structures and equipment operation are all considered training data for this network. Historical runtime sequence data refers to the power equipment operation data continuously collected in chronological order during the historical operation of the target power grid. This type of data completely records the trajectory of the operating status changes of power equipment under different operating conditions and environments. For example, it may be a record of the switching status and operating current of a switchgear in a distribution network under different seasons and load conditions over time.

[0019] It should be noted that the historical interval adjustment amount refers to the interval adjustment amount determined for each operating data of each power device during the historical operation of the target power grid. It is a quantitative value obtained after analyzing and calculating the historical operating data of the power grid. It can reflect the adjustment requirements of the normal interval of the equipment operating data due to changes in the operating status of the power grid under historical conditions. It is an important basis for the spatiotemporal convolutional network to learn the dynamic adjustment benchmark normal interval. For example, during the historical low electricity consumption period of the power grid, the interval adjustment amount determined to correct the normal interval of the voltage operating data of a certain reactor is the historical interval adjustment amount corresponding to that operating data of the power device. The normal interval is determined according to the operating requirements.

[0020] In one possible implementation, the process of determining the historical interval adjustment amount includes: extracting the historical operating data of each power device in the target power grid within a preset time length before the fault occurs from the historical operating sequence data; performing data analysis on the historical operating data of each power device to obtain the data development pattern of the historical operating data of each power device; determining multiple abnormal intervals corresponding to each power device based on the data development pattern; and comparing the multiple abnormal intervals of each power device with the corresponding normal intervals to obtain the multiple historical interval adjustment amounts of each power device.

[0021] In some embodiments, the preset time length is a pre-defined continuous time range tracing back from the time of the fault occurrence. Its duration can be flexibly set according to the type of power grid fault, the operating characteristics of the power equipment, and the needs of fault analysis. Within the preset time length, the operating data of each power device can fully reflect the changing trend of the device's operating data before the fault occurred. Data development patterns refer to the inherent trends, fluctuation characteristics, and change patterns of the equipment's operating data over time, summarized through data analysis of historical operating data of the power equipment. These patterns can clearly reflect the characteristics of data changes when the equipment is operating normally or gradually approaching a fault state. For example, the small fluctuations in the operating voltage of a transmission line with changes in grid load, or the slow upward trend in transformer oil temperature with increasing operating time.

[0022] It should be noted that the abnormal range is determined based on the development patterns of historical operating data of power equipment and the operating state characteristics of power equipment during faults. Before the power equipment fault, the maximum and minimum values ​​of various operating data over time are the abnormal ranges corresponding to those operating data. The abnormal ranges of various power equipment are compared with the corresponding benchmark normal ranges of the equipment. The numerical differences, range offsets, and directions between the abnormal ranges and the benchmark normal ranges are the historical range adjustment amounts. For example, taking a single 10kV oil-immersed distribution transformer as an example, the historical operating data of the top oil temperature within a preset time period of 72 hours before each oil temperature-related fault (such as core overheating or abnormal oil temperature rise caused by winding insulation aging) of the distribution transformer is extracted from the historical operating sequence database. Data analysis of the historical operating data of the top oil temperature is performed using methods such as time series trend fitting, load-oil temperature correlation analysis, and ambient temperature-oil temperature fluctuation analysis. It is found that the transformer top oil temperature has a linear upward trend with the increase of the grid load rate. For each gradient increase in load rate, the oil temperature increases by a certain amount, and the fault... In the 24 hours prior to the incident, the oil temperature deviates from the normal linear rise pattern, exhibiting an abnormally steep increase without load support. Based on the aforementioned data development pattern, the oil temperature data recorded in the 24 hours prior to the incident can be arranged from largest to smallest. Based on the total data volume, an interval can be determined (for example, the minimum and maximum values ​​in the middle 80% of the data can be used as this interval). This interval can be compared with the normal interval, and the difference between the upper limit of this interval and the upper limit of the normal interval, as well as the difference between the lower limit of this interval and the lower limit of the normal interval, can be used as the historical interval adjustment amount.

[0023] Step S120: Compare the real-time operating data in the operating sequence data of each power device with the corresponding baseline normal range to obtain the abnormal operating data of each power device, and determine at least one abnormal propagation path of the target power grid based on the topology and abnormal operating data of each power device.

[0024] In some embodiments, real-time operating data refers to the operating data of power equipment at the current operating moment, which is the last set of data in the operating sequence data. Examples include the instantaneous current value of a high-voltage transmission line, the current real-time oil temperature value of a transformer, and the current operating voltage value of a switchgear. Comparing the real-time value of each operating data item of the power equipment with the corresponding reference normal range allows identification of whether the real-time operating data is within a reasonable normal range. Abnormal operating data refers to the real-time operating data of power equipment that deviates from the corresponding normal range after comparison with the reference normal range. For example, the real-time operating voltage value of a circuit breaker is lower than the lower limit of its reference normal range, or the real-time power value of a transmission line is higher than the upper limit of its reference normal range.

[0025] It should be noted that the anomaly propagation path refers to the path along which the abnormal operating state of equipment in the target power grid spreads from the initially abnormal equipment to other related power equipment along the electrical connection relationship between the power equipment. The anomaly propagation path includes the power equipment that has an electrical connection and shows abnormal operating data, and the electrical connection relationship between these power equipment. For example, after the main transformer of a substation has an anomaly, the anomaly spreads along the electrical connection relationship of the main transformer-overhead transmission line-distribution transformer. This electrical connection line formed by the related abnormal equipment is an anomaly propagation path.

[0026] In one possible implementation, step S120 is specifically processed as follows: for each power device with abnormal operating data, the time of the abnormality of the power device is determined from the operating sequence data; according to the topology of the target power grid, the electrical connection relationship between each power device and other power devices in the target power grid is determined; power devices with electrical connection relationship and the same time of abnormality are identified as power devices on the same abnormality propagation path, and the power devices located on the same propagation path are connected according to the topology relationship to obtain the abnormality propagation path.

[0027] In some embodiments, the anomaly occurrence time refers to the time when the operating data of a power device first deviates from the normal operating data, and the anomaly occurrence time can be extracted from the operating sequence data. For example, if the oil temperature data of a transformer shows an irregular trend before time 1, but shows a continuously rising value after time 1, then time 1 is the anomaly occurrence time of that transformer. Electrical connection relationships refer to the electrical-level associations and pathways between different power devices in the target power grid based on functions such as power transmission, distribution, and transformation. It is a core component of the power grid topology, directly reflecting the specific methods and association logic of power transmission between devices, and is also a key basis for determining the device arrangement order in the anomaly propagation path. For example, the direct connection between a generator and a step-up transformer, the transmission relationship between a distribution transformer and a low-voltage distribution cabinet through transmission lines, and the parallel connection between multiple transmission lines and busbars are all typical electrical connection relationships in a power grid.

[0028] Step S130: Based on the abnormal operation data of all power equipment on each abnormal propagation path, adjust the abnormal operation data of each power equipment on the abnormal propagation path, and determine the abnormal information of each power equipment based on the adjustment results of each power equipment.

[0029] In some embodiments, the adjustment result refers to the final data obtained after performing adjustment operations on the abnormal operating data of the power equipment. If the power equipment is not abnormal, the abnormal operating data of the power equipment is adjusted to normal operating data; if the power equipment is abnormal, the abnormal operating data of the power equipment is not adjusted. The anomaly information is determined based on the adjustment result of the power equipment, the equipment type, and the preset anomaly corresponding rules. The anomaly information includes the anomaly type, the location where the anomaly occurred, and the core related cause of the anomaly.

[0030] See Figure 2 For each abnormal propagation path, the specific processing method of step S130 above includes steps S1301-S1305, the details of which are as follows: Step S1301: Calculate the degree of abnormality of each power device on the propagation path using the abnormal operation data of each power device on the abnormal propagation path and the corresponding baseline normal interval.

[0031] In some embodiments, the degree of anomaly is obtained through quantitative calculation and is used to characterize the extent to which the abnormal operating data of power equipment deviates from the baseline normal range. It can intuitively reflect the differences in the severity of anomalies between different devices. For example, if the degree of anomaly of a certain transmission line is higher than that of a low-voltage distribution cabinet on the same path, it means that the operating data of the transmission line deviates from the normal range more significantly and the anomaly is more severe. If the degree of anomaly of a certain reactor is lower, it means that its data deviates from the normal range less significantly and the anomaly is relatively minor.

[0032] In one possible implementation, step S1301 is specifically processed as follows: for each power device on each abnormal propagation path, calculate the first difference between the abnormal operation data of the power device and the upper limit of the corresponding benchmark normal range, and the second difference between the abnormal operation data of the power device and the lower limit of the corresponding benchmark normal range; determine the degree of abnormality of the power device by the ratio of the minimum of the absolute value of the first difference and the absolute value of the second difference to the corresponding abnormal operation data.

[0033] In some embodiments, the first difference refers to the result obtained by subtracting the abnormal operating data of the power equipment from the upper limit of the corresponding benchmark normal range. The result can be either positive or negative. A positive value indicates that the data exceeds the upper limit, while a negative value indicates that the data does not reach the upper limit. For example, subtracting the upper limit of the benchmark normal range from the abnormal operating data of a transformer oil temperature yields the first difference. The second difference refers to the result obtained by subtracting the abnormal operating data of the power equipment from the lower limit of the corresponding benchmark normal range. The result can also be either positive or negative. A negative value indicates that the data is below the lower limit, while a positive value indicates that the data is above the lower limit. For example, subtracting the lower limit of the benchmark normal range from the abnormal operating data of a reactor operating current yields the second difference.

[0034] In some embodiments, the smaller of the absolute values ​​of the first and second differences (non-negative values) represents the magnitude of the abnormal operating data's deviation from the baseline normal range. This is a key quantitative basis for subsequent calculations of the degree of equipment anomaly. For example, comparing the absolute values ​​of the first and second differences of the current of a power device, the smaller value indicates the degree of anomaly in the current of that power device. Dividing the selected minimum value by the corresponding abnormal operating data of the power device quantifies the abnormal data. This quantifies the relative magnitude of the abnormal data's deviation from the normal range and serves as the direct numerical basis for ultimately determining the degree of anomaly.

[0035] Step S1302: The N power devices with the highest degree of abnormality are identified as candidate abnormal devices, and the abnormality rate of each candidate abnormal device is calculated based on the degree of abnormality of each power device and the positional relationship between each power device and the candidate abnormal devices; where N is a positive integer and N is greater than or equal to 2.

[0036] In some embodiments, candidate abnormal devices refer to the N power devices with the highest degree of abnormality selected from the same abnormal propagation path. Candidate abnormal devices are devices that may have abnormalities, but they may not necessarily have abnormalities. Therefore, further calculations are needed to determine whether they are the real sources of abnormality. For example, if the two power devices with the highest degree of abnormality selected from a certain path are the main transformer and the high-voltage switchgear, then these two devices are the candidate abnormal devices for that abnormal path.

[0037] It should be noted that the positional relationship between power equipment and candidate anomalous equipment refers to the relative electrical connection and spatial arrangement of any power equipment and candidate anomalous equipment in the power grid topology within the anomaly propagation path. For example, if a candidate anomalous equipment is equipment 1, and power equipment 3 is connected to equipment 1 through equipment 2, then the positional relationship between power equipment 3 and equipment 1 is that they are electrically connected but separated by one equipment. The anomaly rate is a quantitative indicator calculated by combining the anomaly degree of all power equipment within the anomaly propagation path and the positional relationship between each power equipment and the candidate anomalous equipment. This indicator is used to characterize the probability that a candidate anomalous equipment becomes the initial source of a power grid anomaly. Its value directly reflects the likelihood that the candidate equipment is the actual source of the anomaly and is the core basis for subsequently determining whether it is the final anomalous equipment. For example, if the anomaly rate value of a candidate anomalous equipment is high, it indicates that it is more likely to be the initial source of the anomaly in that path; if the anomaly rate value is low, it indicates that it is less likely to be the actual source of the anomaly.

[0038] In one possible implementation, step S1302 is specifically processed as follows: based on the positional relationship between each power device and the candidate abnormal device, determine the number of power devices that are spaced between each power device and the candidate abnormal device; for each candidate abnormal device, perform the following steps: plot an abnormality curve with the number of power devices that are spaced between each power device and the candidate abnormal device as the horizontal axis and the abnormality degree of each power device as the vertical axis; calculate the similarity between the abnormality degree curve and the standard abnormality propagation curve, and determine the similarity as the abnormality rate of the candidate abnormal device.

[0039] In some embodiments, the anomaly degree curve is a visually generated curve that can intuitively reflect the anomaly degree of each power device within the anomaly propagation path. As the number of power devices between the power device and the candidate anomaly device changes, the anomaly degree curve will show a certain trend. For example, a certain curve shows a trend of gradually decreasing anomaly degree as the number of intervals increases, intuitively reflecting the characteristic that the farther away from the candidate anomaly device, the milder the device anomaly.

[0040] The standard anomaly propagation curve is a typical trend curve obtained by fitting data from objective laws governing power grid operation and a large amount of historical fault data. It reflects the inherent law that the degree of equipment anomaly changes with the distance from the anomaly source as an anomaly propagates from its initial source to surrounding equipment. It is the core reference benchmark for determining whether a candidate anomaly device is a true anomaly source. For example, when an anomaly in the power grid propagates from its source, the degree of anomaly usually decreases with distance from the source. The trend curve fitted based on this is the standard anomaly propagation curve. Similarity refers to the degree of similarity between the plotted anomaly degree curve and the standard anomaly propagation curve. The similarity can be calculated using algorithms for curve similarity. For example, if the shape and trend of an anomaly degree curve are almost identical to the standard anomaly propagation curve, the similarity value is high, and the anomaly rate of the corresponding candidate anomaly device is also high.

[0041] Step S1303: Compare the abnormality rate of each candidate abnormal device with a preset threshold, and determine at least one abnormal device on the abnormal propagation path based on the comparison result of each candidate abnormal device.

[0042] In some embodiments, the preset threshold is pre-set and can be set according to actual conditions. When the failure rate of candidate abnormal devices exceeds the preset threshold, the candidate abnormal devices with failure rates exceeding the preset threshold need to be identified as abnormal devices. When the failure rate of no candidate abnormal devices exceeds the preset threshold, the candidate abnormal device with the highest failure rate needs to be identified as an abnormal device.

[0043] Step S1304: Adjust the abnormal operation data of each power device on the abnormal propagation path according to the positional relationship between each power device on the abnormal propagation path and at least one abnormal device.

[0044] It should be noted that the abnormal operating data of electrical equipment along the anomaly propagation path may be due to the propagation of an abnormality from another abnormal electrical device, or it may be due to an inherent anomaly within the device itself. If the abnormal operating data of a particular electrical device is caused by the propagation of an abnormality from another abnormal device, then the abnormal operating data of that device needs to be adjusted to normal operating data. If the abnormal operating data of a particular electrical device is caused by an inherent anomaly within the device itself, then that device needs to be identified as an abnormal electrical device. For example, if a phase-to-phase short-circuit fault occurs on the 10kV high-voltage busbar of a city's distribution network, the fault anomaly will propagate along the electrical connections of the grid topology, forming a fixed anomaly propagation path: high-voltage busbar - incoming circuit breaker - 10kV transmission line - distribution transformer - low-voltage switchgear. During the fault propagation process, it exhibits a directional propagation along the topological path, with the degree of anomaly decreasing as the number of intervals increases, and the proportion of false anomalies increasing as the number of intervals increases. At this time, the incoming circuit breaker directly connected to the high-voltage busbar is most affected, and its current operating data significantly exceeds the baseline normal range; the current data of the transmission line exceeds the baseline normal range by a smaller margin than that of the incoming circuit breaker; the voltage data of the distribution transformer only slightly exceeds the baseline normal range, which is a data anomaly caused by fault propagation; the voltage data of the low-voltage distribution cabinet only shows small fluctuations.

[0045] In one possible implementation, step S1304 is specifically processed as follows: For each power device, the following steps are performed: If there is an abnormal device on the anomaly propagation path, the number of power devices between the power device and the abnormal device is determined according to the positional relationship between the power device and the abnormal device, and an anomaly severity threshold is determined based on the number. When the anomaly severity of the power device is less than the anomaly severity threshold, the abnormal operation data of the power device is adjusted to normal operation data. When the anomaly severity of the power device is not less than the anomaly severity threshold, the power device is identified as an abnormal device. If there are multiple abnormal devices on the anomaly propagation path, the minimum number of power devices between the power device and each abnormal device is determined according to the positional relationship between the power device and each abnormal device. An anomaly severity threshold is determined based on the minimum number. When the anomaly severity of the power device is less than the anomaly severity threshold, the abnormal operation data of the power device is adjusted to normal operation data. When the anomaly severity of the power device is not less than the anomaly severity threshold, the power device is identified as an abnormal device.

[0046] In some embodiments, the anomaly severity threshold is a critical value determined based on the number of intervals between the power equipment and the nearest anomalous equipment. It is the core standard for determining whether the anomalous power equipment is a genuine anomaly caused by its own fault or a false anomaly caused by the propagation of the anomaly. Different numbers of intervals correspond to different anomaly severity thresholds. For example, for transmission line-transformer propagation paths in urban power distribution networks, the anomaly severity threshold can be preset to 0.3 when the number of intervals is 0, based on historical data. Since electrical signal propagation attenuation is relatively slow in urban power distribution networks, the anomaly severity threshold can be reduced by 0.05 for every 1 increase in the number of intervals. Then, based on the number of intervals, the preset anomaly severity threshold when the number of intervals is 0, and the attenuation rate of the anomaly severity threshold, the anomaly severity threshold on this propagation path can be calculated. In this case, if the number of intervals is 1, the anomaly severity threshold is 0.3 - 0.05 × 1 = 0.25. It should be noted that different anomaly severity thresholds corresponding to different numbers of intervals need to be determined based on historical operating data, and different anomalous equipment and different anomalies may correspond to different anomaly severity thresholds.

[0047] It should be noted that normal operating data refers to the operating data of power equipment that is within the corresponding baseline normal range and reflects the normal working status of the equipment. When adjusting the abnormal operating data of power equipment caused by abnormal power equipment, this abnormal operating data can be adjusted to the operating data of the power equipment before the abnormality.

[0048] Step S1305: For each power device, if the power device is abnormal, then determine the abnormal information of the power device according to the device type, abnormal operation data and the preset abnormal information correspondence table.

[0049] In some embodiments, if a power device itself is malfunctioning, its malfunction information needs to be obtained by combining the device type, abnormal operating data, and an anomaly information correspondence table. The anomaly information correspondence table is a pre-defined standardized table that includes the relationship between different device types for each type of power device, abnormal operating data (the data type and specific values ​​of the abnormal operating data; for example, abnormal operating data includes current and voltage data, with specific values ​​a and b), and corresponding specific anomaly information. The anomaly information may include the specific anomaly type of the power device, the key operating data indicators that triggered the anomaly, the specific location of the anomaly on the device, and, if the power device exhibits different data change trends when malfunctioning for different reasons, the anomaly information may also include the core triggering factor for the anomaly.

[0050] By inputting the target power grid topology and runtime sequence data into a pre-trained spatiotemporal convolutional network, and combining historical interval adjustments, a baseline normal range adapted to the dynamic operating conditions of the target power grid is dynamically generated, avoiding the shortcomings of fixed ranges that cannot adapt to the dynamic operating state of the power grid. By determining the anomaly propagation path and setting an anomaly severity threshold based on the number of intervals between devices and anomaly sources, it is possible to accurately distinguish between data anomalies caused by device faults and data anomalies caused by anomaly propagation. This enables precise location of abnormal devices in the power grid, effectively improving the accuracy and efficiency of power equipment anomaly identification, reducing power grid operation and maintenance costs, and ensuring the stability and security of power grid operation.

[0051] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0052] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0053] Figure 3 The diagram illustrates the structure of a power equipment anomaly identification device based on a spatiotemporal convolutional network according to an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 3 As shown, the power equipment anomaly identification device 3 based on spatiotemporal convolutional networks includes: The determination module 31 is used to input the topology of the target power grid and the runtime sequence data within the preset time window into the pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each running data of each power device, and determine the benchmark normal interval based on the interval adjustment amount. The comparison module 32 is used to compare the real-time operating data in the operating sequence data of each power device with the corresponding baseline normal range to obtain the abnormal operating data of each power device, and to determine at least one abnormal propagation path of the target power grid based on the topology and abnormal operating data of each power device. The adjustment module 33 is used to adjust the abnormal operation data of each power device on each abnormal propagation path according to the abnormal operation data of all power devices on each abnormal propagation path, and determine the abnormal information of each power device based on the adjustment result of each power device.

[0054] In one possible implementation, the training data for the pre-trained spatiotemporal convolutional network includes the topology of the target power grid, the historical runtime sequence data of the target power grid, and the historical interval adjustment amount corresponding to each operational data of each power device in the target power grid.

[0055] In one possible implementation, the process of determining the historical interval adjustment amount includes: extracting the historical operating data of each power device in the target power grid within a preset time length before the fault occurs from the historical operating sequence data; performing data analysis on the historical operating data of each power device to obtain the data development pattern of the historical operating data of each power device; determining multiple abnormal intervals corresponding to each power device based on the data development pattern; and comparing the multiple abnormal intervals of each power device with the corresponding benchmark normal intervals to obtain the multiple historical interval adjustment amounts of each power device.

[0056] In one possible implementation, the comparison module 32 is specifically used to: determine the time of the abnormality of each power device with abnormal operating data from the runtime sequence data; determine the electrical connection relationship between each power device and other power devices in the target power grid according to the topology of the target power grid; identify power devices with electrical connection relationship and the same time of abnormality as power devices on the same abnormality propagation path, and connect the power devices located on the same abnormality propagation path according to the topology relationship to obtain the abnormality propagation path.

[0057] In one possible implementation, the adjustment module 33 is specifically used to: for each anomaly propagation path, perform the following steps: using the abnormal operation data of each power device on the anomaly propagation path and the corresponding baseline normal interval, calculate the anomaly degree of each power device on the propagation path; determine the N power devices with the highest anomaly degree as candidate anomaly devices, and calculate the anomaly rate of each candidate anomaly device based on the anomaly degree of each power device and the positional relationship between each power device and the candidate anomaly devices; where N is a positive integer and N is greater than or equal to 2; compare the anomaly rate of each candidate anomaly device with a preset threshold, and determine at least one anomaly device on the anomaly propagation path based on the comparison result of each candidate anomaly device; adjust the abnormal operation data of each power device on the anomaly propagation path based on the positional relationship between each power device on the anomaly propagation path and at least one anomaly device; for each power device, if the power device is an anomaly device, determine the anomaly information of the power device based on the device type, abnormal operation data and a preset anomaly information correspondence table.

[0058] In one possible implementation, the adjustment module 33 is further configured to: for each power device, perform the following steps: if there is an abnormal device on the anomaly propagation path, determine the number of power devices between the power device and the abnormal device based on the positional relationship between the power device and the abnormal device, and determine an anomaly severity threshold based on the number; when the anomaly severity of the power device is less than the anomaly severity threshold, adjust the abnormal operation data of the power device to normal operation data; when the anomaly severity of the power device is not less than the anomaly severity threshold, identify the power device as an abnormal device; if there are multiple abnormal devices on the anomaly propagation path, determine the minimum number of power devices between the power device and each abnormal device based on the positional relationship between the power device and each abnormal device; determine the anomaly severity threshold based on the minimum number; when the anomaly severity of the power device is less than the anomaly severity threshold, adjust the abnormal operation data of the power device to normal operation data; when the anomaly severity of the power device is not less than the anomaly severity threshold, identify the power device as an abnormal device.

[0059] In one possible implementation, the adjustment module 33 is further configured to: determine the number of power devices between each power device and the candidate abnormal device based on the positional relationship between each power device and the candidate abnormal device; for each candidate abnormal device, perform the following steps: plot an abnormality curve with the number of power devices between each power device and the candidate abnormal device as the horizontal axis and the abnormality degree of each power device as the vertical axis; calculate the similarity between the abnormality degree curve and the standard abnormality propagation curve, and determine the similarity as the abnormality rate of the candidate abnormal device.

[0060] In one possible implementation, the adjustment module 33 is further configured to: for each power device on each abnormal propagation path, calculate a first difference between the abnormal operating data of the power device and the upper limit of the corresponding benchmark normal range, and a second difference between the abnormal operating data of the power device and the lower limit of the corresponding benchmark normal range; and determine the degree of abnormality of the power device by the ratio of the minimum of the absolute value of the first difference and the absolute value of the second difference to the corresponding abnormal operating data.

[0061] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 4 As shown, the electronic device 4 in this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. When the processor 40 executes the computer program 42, it implements the steps in the various method embodiments described above. Alternatively, when the processor 40 executes the computer program 42, it implements the functions of each module / unit in the various device embodiments described above.

[0062] For example, computer program 42 may be divided into one or more modules / units, which are stored in memory 41 and executed by processor 40 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 42 in electronic device 4.

[0063] Electronic device 4 may include, but is not limited to, processor 40 and memory 41. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 4 may also include input / output devices, network access devices, buses, etc.

[0064] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0065] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0066] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying power equipment anomalies based on spatiotemporal convolutional networks, characterized in that, include: The topology of the target power grid and the runtime sequence data within a preset time window are input into a pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each operational data of each power device, and the benchmark normal interval is determined based on the interval adjustment amount. The real-time operating data in the runtime sequence data of each power device is compared with the corresponding baseline normal range to obtain the abnormal operating data of each power device. Based on the topology and abnormal operating data of each power device, at least one abnormal propagation path of the target power grid is determined. Based on the abnormal operation data of all power equipment along each abnormal propagation path, the abnormal operation data of each power equipment along the abnormal propagation path is adjusted, and the abnormal information of each power equipment is determined based on the adjustment results of each power equipment.

2. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 1, characterized in that, The step of adjusting the abnormal operation data of each power device along each abnormal propagation path based on the abnormal operation data of all power devices along that path, and determining the abnormal information of each power device based on the adjustment results, includes: For each abnormal propagation path, perform the following steps: Using the abnormal operation data of each power device along the abnormal propagation path and the corresponding baseline normal range, the degree of abnormality of each power device along the propagation path is calculated. The N power devices with the highest degree of abnormality are identified as candidate abnormal devices. Based on the degree of abnormality of each power device and the positional relationship between each power device and the candidate abnormal devices, the abnormality rate of each candidate abnormal device is calculated. Wherein, N is a positive integer and N is greater than or equal to 2. The anomaly rate of each candidate abnormal device is compared with a preset threshold, and based on the comparison result of each candidate abnormal device, at least one abnormal device on the anomaly propagation path is determined. Based on the positional relationship between each power device on the anomaly propagation path and the at least one abnormal device, the abnormal operation data of each power device on the anomaly propagation path is adjusted; For each power device, if the power device is abnormal, the abnormal information of the power device is determined according to the device type, abnormal operation data and the preset abnormal information correspondence table.

3. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 2, characterized in that, The step of adjusting the abnormal operation data of each power device along the abnormal propagation path based on the positional relationship between each power device and the at least one abnormal device includes: For each electrical device, perform the following steps: If there is an abnormal device on the abnormal propagation path, the number of power devices between the power device and the abnormal device is determined according to the positional relationship between the power device and the abnormal device, and the abnormality threshold is determined according to the number. When the abnormality of the power device is less than the abnormality threshold, the abnormal operation data of the power device is adjusted to normal operation data. When the abnormality of the power device is not less than the abnormality threshold, the power device is identified as an abnormal device. If there are multiple abnormal devices on the abnormal propagation path, then the minimum number of power devices between the power device and each abnormal device is determined based on the positional relationship between the power device and each abnormal device. The abnormality threshold is determined based on the minimum quantity. When the abnormality of the power equipment is less than the abnormality threshold, the abnormal operation data of the power equipment is adjusted to normal operation data. When the abnormality of the power equipment is not less than the abnormality threshold, the power equipment is identified as abnormal equipment.

4. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 2, characterized in that, The calculation of the abnormality rate of each candidate abnormal device based on the abnormality degree of each power device and the positional relationship between each power device and the candidate abnormal devices includes: Based on the positional relationship between each power device and the candidate abnormal device, determine the number of power devices that are spaced between each power device and the candidate abnormal device; For each candidate faulty device, perform the following steps: Plot an anomaly degree curve with the number of power devices between each power device and the candidate abnormal device as the horizontal axis and the anomaly degree of each power device as the vertical axis. Calculate the similarity between the anomaly degree curve and the standard anomaly propagation curve, and determine the similarity as the anomaly rate of the candidate anomaly device.

5. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 2, characterized in that, The step of calculating the degree of abnormality for each power device along the propagation path using abnormal operation data and the corresponding baseline normal range for each abnormal operation data includes: For each power device on each abnormal propagation path, calculate the first difference between the abnormal operation data of the power device and the upper limit of the corresponding benchmark normal range, and the second difference between the abnormal operation data of the power device and the lower limit of the corresponding benchmark normal range. The ratio of the minimum of the absolute values ​​of the first and second differences to the corresponding abnormal operating data is used to determine the degree of abnormality of the power equipment.

6. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 1, characterized in that, The determination of at least one anomaly propagation path for the target power grid based on the topology and abnormal operation data of each power device includes: For each power device with abnormal operating data, determine the time of occurrence of the abnormality from the runtime sequence data; Based on the topology of the target power grid, determine the electrical connection relationships between each power device and other power devices in the target power grid; Electrical devices that are electrically connected and whose anomalies occur at the same time are identified as electrical devices on the same anomaly propagation path. Based on the topological relationship, the electrical devices located on the same anomaly propagation path are connected to obtain the anomaly propagation path.

7. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 1, characterized in that, The training data of the pre-trained spatiotemporal convolutional network includes the topology of the target power grid, the historical runtime sequence data of the target power grid, and the historical interval adjustment amount corresponding to each operational data of each power device in the target power grid.

8. The power equipment anomaly identification method based on spatiotemporal convolutional networks according to claim 1, characterized in that, The process for determining the historical interval adjustment amount includes: Extract the historical operating data of each power device in the target power grid within a preset time period before the fault occurs from the historical operating sequence data; Data analysis is performed on the historical operating data of each power device to obtain the data development patterns of the historical operating data of each power device. Based on the aforementioned data development patterns, multiple abnormal intervals were identified for each power device. By comparing multiple abnormal intervals of each power device with the corresponding baseline normal interval, the adjustment amount of multiple historical intervals for each power device is obtained.

9. A power equipment anomaly identification device based on spatiotemporal convolutional networks, characterized in that, include: The determination module is used to input the topology of the target power grid and the runtime sequence data within a preset time window into a pre-trained spatiotemporal convolutional network to obtain the interval adjustment amount corresponding to each running data of each power device, and determine the benchmark normal interval based on the interval adjustment amount. The comparison module is used to compare the real-time operating data in the runtime sequence data of each power device with the corresponding baseline normal range to obtain the abnormal operating data of each power device, and to determine at least one abnormal propagation path of the target power grid based on the topology and abnormal operating data of each power device. The adjustment module is used to adjust the abnormal operation data of each power device on each abnormal propagation path based on the abnormal operation data of all power devices on each abnormal propagation path, and determine the abnormal information of each power device based on the adjustment results of each power device.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 8.

Images