Reference maintenance method, device and equipment for power grid operation state switching

Abstract

The embodiment of the invention provides a reference maintenance method, device and equipment for power grid operation state switching, and is applied to the technical field of power system monitoring. The method comprises the steps of continuously collecting multi-source heterogeneous data of a target power grid in a current time period, and performing unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain an available data set; performing operation state verification on the target power grid based on the available data set, and when a verification result indicates that the target power grid is subjected to working condition switching, extracting first real-time qualified data after working condition switching from the available data set; based on the first real-time qualified data, constructing a shadow benchmark of the target power grid, and executing playback verification on the shadow benchmark by adopting accumulated third real-time qualified data after working condition switching to generate a playback verification result; and if the playback verification result meets the preset switching condition, updating the operation reference currently used by the target power grid to be the shadow reference. The technical effect of improving the risk inspection accuracy is achieved.

Description

Technical Field

[0001] This application relates to the field of power system monitoring technology, and in particular to a benchmark maintenance method, device and equipment for switching power grid operating states. Background Technology

[0002] With the continuous expansion of the power grid and the rapid development of inter-regional power transmission, power grid operation monitoring data now encompasses multiple heterogeneous sources. Since the quality of this multi-source heterogeneous data directly impacts the safe operation of the power grid, data auditing is necessary to ensure the stability and security of the power grid under complex operating environments.

[0003] In existing technologies, data auditing methods mainly rely on local neighborhood consistency or statistical deviation discrimination techniques. Specifically, algorithms based on impedance distance, phasor similarity, or distribution offset identify anomalies by comparing the statistical characteristics of local data points.

[0004] Because existing technologies cannot continuously track risks in power grid monitoring data, there is a technical problem of low accuracy in risk auditing. Summary of the Invention

[0005] This application provides a benchmark maintenance method, apparatus, and equipment for switching power grid operating states, in order to improve the accuracy of risk auditing.

[0006] In a first aspect, embodiments of this application provide a baseline maintenance method for switching power grid operating states, including:

[0007] Continuously collect multi-source heterogeneous data of the target power grid within the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset;

[0008] The target power grid is verified based on the available dataset. When the verification result of the target power grid's operating status indicates that the target power grid has undergone a change in operating condition, the first real-time qualified data after the change in operating condition is extracted from the available dataset.

[0009] Based on the first real-time qualified data after the operating condition switch, a shadow benchmark of the target power grid is constructed, and the shadow benchmark is replayed and verified using the accumulated third real-time qualified data after the operating condition switch in the available data set, and the replay verification result is generated.

[0010] If the playback verification result meets the preset switching conditions, the operating reference currently used by the target power grid will be updated to the shadow reference; where the operating reference refers to the reference standard adapted to the operating conditions of the target power grid.

[0011] In one possible implementation, multi-source heterogeneous data of the target power grid is continuously collected within the current time period, and the multi-source heterogeneous data is aligned to a unified time standard and filtered for quality to obtain a usable dataset, including:

[0012] Continuously collect multi-source heterogeneous data of the target power grid during the current period; the multi-source heterogeneous data includes core power grid data, meteorological data, and planning data;

[0013] Based on multiple data sources corresponding to multi-source heterogeneous data, a unified time base is established for multiple data sources;

[0014] The raw data from multiple data sources are matched according to time slices to obtain multi-source heterogeneous data under different time slices;

[0015] For each time slice, a quality check is performed on the multi-source heterogeneous data. The multi-source heterogeneous data that passes the check is retained and combined according to the time order of the time slice to obtain a usable dataset.

[0016] In one possible implementation, the target power grid's operational status is verified based on the available dataset. When the verification result indicates a change in operating condition, the first real-time qualified data after the change in operating condition is extracted from the available dataset, including:

[0017] Based on the available dataset, obtain second real-time qualified data for multiple consecutive sampling periods;

[0018] When the second real-time qualified data indicates a change in the status of the circuit breaker in the target power grid, or when the rate of change of the injected power of the power node in the target power grid reaches a preset rate of change threshold, it is determined that the operating condition of the target power grid has changed.

[0019] Based on the available dataset, obtain the first real-time qualified data after the operating condition switch.

[0020] In one possible implementation, a shadow benchmark of the target power grid is constructed based on the first real-time qualified data after the operating condition switch, including:

[0021] Based on a preset fixed-length sliding window and a preset fixed step size, statistical modeling is performed on the first real-time qualified data to obtain a shadow benchmark.

[0022] In one possible implementation, the shadow baseline is replayed and verified using the accumulated third real-time qualified data after the switching of operating conditions in the available dataset, generating a replay verification result, including:

[0023] Extract the third real-time qualified data accumulated after the operating condition switch from the available dataset;

[0024] The third real-time qualified data is input into the shadow benchmark. The shadow benchmark is used to determine whether multiple multi-source heterogeneous data in the third real-time qualified data belong to the normal operating state, and multiple sets of discrimination results are obtained.

[0025] The false alarm rate and false negative rate of the shadow benchmark are calculated based on the discrimination results, and the playback verification results are generated.

[0026] In one possible implementation, the preset switching conditions include:

[0027] The false alarm rate in the playback verification results did not exceed the preset false alarm rate, and the false detection rate in the playback verification results did not exceed the preset false detection rate.

[0028] In one possible implementation, the method further includes, prior to performing an operational status verification of the target power grid based on the available dataset:

[0029] Extract core elements from available datasets, including power grid core data, meteorological data, and planning data, and construct a cross-domain heterogeneous graph.

[0030] Based on preset core physical rules, pruning is performed on cross-domain heterogeneous graphs to delete edges that violate physical constraints;

[0031] The pruned cross-domain heterogeneous graph is used as the coupling graph of the target power grid;

[0032] The coupling graph is used for the subsequent calculation of risk quantification indicators.

[0033] In one possible implementation, after updating the operating reference currently used by the target power grid to a shadow reference, the method further includes:

[0034] Extract the fourth real-time qualified data after the baseline switch is completed from the available dataset;

[0035] Based on the coupling diagram of the target power grid and the fourth set of qualified real-time data, the index values ​​of three sets of risk quantification indicators are calculated; the risk quantification indicators refer to the low-frequency energy increment, feasibility gap, and data source credibility.

[0036] Based on the benchmark currently used by the target power grid, obtain the benchmark distribution corresponding to the benchmark;

[0037] Based on the indicator values ​​of the three sets of risk quantification indicators, a joint distribution of the indicators corresponding to the indicator values ​​is generated.

[0038] The distribution deviation value is calculated based on the joint distribution of indicators and the benchmark distribution;

[0039] Based on the distribution deviation value and the index values ​​of three sets of risk quantification indicators, the audit risk score of the target power grid is calculated.

[0040] Secondly, embodiments of this application provide a reference maintenance device for switching power grid operating states, comprising:

[0041] The acquisition module is used to continuously collect multi-source heterogeneous data of the target power grid within the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset;

[0042] The first processing module is used to verify the operating status of the target power grid based on the available dataset. When the verification result of the operating status of the target power grid indicates that the target power grid has undergone a change in operating condition, the first real-time qualified data after the change in operating condition is extracted from the available dataset.

[0043] The second processing module is used to construct a shadow benchmark of the target power grid based on the first real-time qualified data after the operating condition switch, and to perform a replay verification on the shadow benchmark using the accumulated third real-time qualified data after the operating condition switch in the available data, and generate a replay verification result.

[0044] The third processing module is used to update the current operating reference of the target power grid to the shadow reference when the playback verification result meets the preset switching conditions.

[0045] In one possible implementation, the acquisition module is further configured to:

[0046] Continuously collect multi-source heterogeneous data of the target power grid during the current period; the multi-source heterogeneous data includes core power grid data, meteorological data, and planning data;

[0047] Based on multiple data sources corresponding to multi-source heterogeneous data, a unified time base is established for multiple data sources;

[0048] The raw data from multiple data sources are matched according to time slices to obtain multi-source heterogeneous data under different time slices;

[0049] For each time slice, a quality check is performed on the multi-source heterogeneous data. The multi-source heterogeneous data that passes the check is retained and combined according to the time order of the time slice to obtain a usable dataset.

[0050] In one possible implementation, the first processing module is further configured to:

[0051] Based on the available dataset, obtain second real-time qualified data for multiple consecutive sampling periods;

[0052] When the second real-time qualified data indicates a change in the status of the circuit breaker in the target power grid, or when the rate of change of the injected power of the power node in the target power grid reaches a preset rate of change threshold, it is determined that the operating condition of the target power grid has changed.

[0053] Based on the available dataset, obtain the first real-time qualified data after the operating condition switch.

[0054] In one possible implementation, the second processing module is further configured to:

[0055] Based on a preset fixed-length sliding window and a preset fixed step size, statistical modeling is performed on the first real-time qualified data to obtain a shadow benchmark.

[0056] In one possible implementation, the shadow baseline is replayed and verified using the accumulated third real-time qualified data after the switching of operating conditions in the available dataset, generating a replay verification result, including:

[0057] Extract the third real-time qualified data accumulated after the operating condition switch from the available dataset;

[0058] The third real-time qualified data is input into the shadow benchmark. The shadow benchmark is used to determine whether multiple multi-source heterogeneous data in the third real-time qualified data belong to the normal operating state, and multiple sets of discrimination results are obtained.

[0059] The false alarm rate and false negative rate of the shadow benchmark are calculated based on the discrimination results, and the playback verification results are generated.

[0060] In one possible implementation, the preset switching conditions include:

[0061] The false alarm rate in the playback verification results did not exceed the preset false alarm rate, and the false detection rate in the playback verification results did not exceed the preset false detection rate.

[0062] In one possible implementation, the acquisition module is further configured to:

[0063] Extract core elements from available datasets, including power grid core data, meteorological data, and planning data, and construct a cross-domain heterogeneous graph.

[0064] Based on preset core physical rules, pruning is performed on cross-domain heterogeneous graphs to delete edges that violate physical constraints;

[0065] The pruned cross-domain heterogeneous graph is used as the coupling graph of the target power grid;

[0066] The coupling graph is used for the subsequent calculation of risk quantification indicators.

[0067] In one possible implementation, the third processing module is further configured to:

[0068] Extract the fourth real-time qualified data after the baseline switch is completed from the available dataset;

[0069] Based on the coupling diagram of the target power grid and the fourth set of qualified real-time data, the index values ​​of three sets of risk quantification indicators are calculated; the risk quantification indicators refer to the low-frequency energy increment, feasibility gap, and data source credibility.

[0070] Based on the benchmark currently used by the target power grid, obtain the benchmark distribution corresponding to the benchmark;

[0071] Based on the indicator values ​​of the three sets of risk quantification indicators, a joint distribution of the indicators corresponding to the indicator values ​​is generated.

[0072] The distribution deviation value is calculated based on the joint distribution of indicators and the benchmark distribution;

[0073] Based on the distribution deviation value and the index values ​​of three sets of risk quantification indicators, the audit risk score of the target power grid is calculated.

[0074] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0075] The memory stores instructions that the computer executes;

[0076] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect above and various possible implementations of the first aspect.

[0077] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and various possible implementations thereof.

[0078] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and various possible implementations thereof.

[0079] This application provides a method, apparatus, and equipment for maintaining a reference for power grid operation state switching. The method continuously collects multi-source heterogeneous data of the target power grid within the current time period, performs unified time standard and quality screening processing on the multi-source heterogeneous data, and obtains a usable dataset. Based on the usable dataset, it detects whether the target power grid has undergone an operation state switch. When an operation state switch is determined to have occurred, it extracts the first real-time qualified data after the operation state switch from the usable dataset and uses the first real-time qualified data to construct a shadow reference for the target power grid. When the playback verification result of the shadow reference is detected to meet preset switching conditions, the reference currently used by the target power grid is replaced with the shadow reference. Compared with the prior art, when it is determined that the operation state of the target power grid has switched, this application first constructs a shadow reference that meets the operating state after the switch based on continuously collected data, while maintaining the current reference; when it is confirmed that the shadow reference meets the switching conditions, the current reference is switched to the shadow reference. This eliminates the need to wait for threshold hysteresis adjustment and allows for continuous auditing of multi-source heterogeneous data, thereby avoiding the technical problem of low accuracy in risk auditing due to the inability to audit some data, and achieving the technical effect of improving the accuracy of risk auditing. Attached Figure Description

[0080] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0081] Figure 1 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 1 ;

[0082] Figure 2 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 2 ;

[0083] Figure 3 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 3 ;

[0084] Figure 4 A flowchart illustrating the power grid risk auditing method provided in this application embodiment;

[0085] Figure 5 A schematic diagram of the structure of the reference maintenance device for switching power grid operating states provided in this application;

[0086] Figure 6 A schematic diagram of the structure of the electronic device provided in this application.

[0087] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0088] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0089] First, the terms used in this application will be explained:

[0090] A phase measurement unit (PMU) is a high-precision measurement device used in power grids to measure voltage phase angle, vector current, and system frequency.

[0091] Supervisory Control and Data Acquisition (SCADA) system refers to a power grid dispatching and monitoring system used to collect data such as voltage, current, power, and switch tag numbers from the power grid. RTU refers to the remote terminal unit of SCADA.

[0092] Automatic Generation Control (AGC) refers to the frequency or power regulation system of the power grid dispatching system, which is used to maintain the frequency stability of the power grid, track the power generation plan of the power grid, and control the power balance between the power input and output of the power grid.

[0093] Network Time Protocol (NTP) is a network protocol used to unify the time base of multiple devices.

[0094] Precision Time Protocol (PTP) is a protocol used for high-precision time synchronization among multiple devices.

[0095] The Autoregressive Integrated Moving Average (ARIMA) model is a statistical model used for time series forecasting and modeling, designed to fit the distribution patterns of time series data.

[0096] Gaussian Mixture Model (GMM) is a tool used for nonparametric statistical modeling. It can describe complex multimodal data distributions by fitting the parameters of multiple Gaussian distributions.

[0097] Wasserstein distance is a metric that measures the difference between two probability distributions. It is calculated by determining the minimum cost required to transform one distribution into another.

[0098] Jensen-Shannon distance: refers to a distance value obtained by symmetric and normalizing the divergence of two distributions, and its range is between 0 and 1.

[0099] In existing technologies, the main approach to risk auditing of power grid operational data is to use algorithms based on impedance distance, phasor similarity, or distribution offset to identify anomalies by comparing the statistical characteristics of local data points. This allows for the identification of abnormal data and further risk auditing. However, since power grid topology and scheduling plans may change, the auditing benchmarks corresponding to the data in the grid may no longer be applicable to the changed data. Therefore, it is necessary to temporarily interrupt the process using a freeze window or a dual-threshold hysteresis strategy, waiting for the generation of new auditing benchmarks before conducting further risk auditing.

[0100] However, due to existing data auditing methods, during the switching of power grid operation status, existing technologies may have blind spots in risk monitoring due to freezing windows or dual-threshold hysteresis strategies, making it impossible to continuously track risks. As a result, some data in existing technologies cannot be audited, leading to the technical problem of low accuracy in risk auditing.

[0101] To address the aforementioned technical issues, this application proposes the following technical concept: by constructing a unified processing framework for multi-source heterogeneous data, continuous tracking of power grid operation risks is achieved through dynamic benchmark maintenance. Specifically, this technical concept is based on unified time scales and quality screening, and solves the monitoring blind spot problem during operating condition switching by real-time identification of operating state transitions and smooth migration of benchmark models, thereby achieving the technical effect of improving the accuracy of risk auditing.

[0102] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0103] Figure 1 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 1 ,like Figure 1 As shown, the method includes:

[0104] S101. Continuously collect multi-source heterogeneous data of the target power grid in the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset.

[0105] In this step, the construction of the usable dataset refers to the process of transforming the multi-source heterogeneous raw data of the target power grid into a standardized dataset that is time-consistent, of appropriate quality, and can be directly used for subsequent analysis, through a unified time base, matching time slices, and then quality screening. Furthermore, the usable dataset is continuously updated, retaining data over a period of time that includes real-time data at the current moment.

[0106] Alternatively, one possible implementation for constructing the available dataset is as follows:

[0107] S1011. Continuously collect multi-source heterogeneous data of the target power grid during the current time period.

[0108] In this step, the multi-source heterogeneous data includes core power grid data, meteorological data, and planning data. Core power grid data refers to PMU data, SCADA data, and topology status data of the target power grid. PMU data includes, but is not limited to, voltage phase angles, vector currents, and frequencies of each electrical node in the target power grid. SCADA data includes, but is not limited to, bus voltages, line currents, active power, reactive power, and circuit breaker tripping by switch tag number of the target power grid. Topology status data includes, but is not limited to, the connection relationships between buses and lines in the target power grid, and the operational and decommissioning status of equipment.

[0109] Meteorological data refers to the transmission meteorological monitoring data of the target power grid, including but not limited to wind speed, wind direction, ambient temperature, icing thickness, lightning density, and solar irradiance along the transmission line corridor.

[0110] Planned data refers to the dispatch planning domain data of the target power grid, including but not limited to the unit handling plan, AGC setpoint, planned power flow data, and tie line planned input and output capacity of the target power grid.

[0111] S1012. Based on multiple data sources corresponding to multi-source heterogeneous data, unify the time base of multiple data sources.

[0112] In this step, unifying the time base refers to consistently calibrating all data sources using NTP and PTP based on the master clock of the target power grid dispatch center, eliminating clock drift of individual devices, and ensuring that the timestamps of all data are based on the same clock base. The data sources can be: PMU devices, SCADA RTU terminals, weather stations, and dispatch system servers.

[0113] S1013. Match the raw data from multiple data sources according to time slices to obtain multi-source heterogeneous data under different time slices.

[0114] In this step, time-slice matching refers to matching the calibrated raw data according to the time slices. Each time slice corresponds to a set of multi-source data at the same time, thus forming a correspondence between time slices and multi-source heterogeneous data.

[0115] S1014. Perform quality checks on the multi-source heterogeneous data under each time slice, retain the multi-source heterogeneous data that passes the checks, and combine them according to the time order of the time slices to obtain a usable dataset.

[0116] In this step, quality checks include time consistency checks, missing rate checks, value range checks, limit violation checks, and dimensional consistency checks. Specifically, time consistency checks verify that the clock deviation of raw data from different sources is below a corresponding preset deviation value. Missing rate checks verify that the exact proportion of data in a single category is below a preset missing rate threshold. Value range checks verify that the data is within a preset reasonable range. Limit violation checks verify that the data exceeds the equipment's safe operating threshold. Dimensional consistency checks verify that the units of the data conform to preset standard units.

[0117] For example, a time consistency check can be performed to determine whether the PMU data clock deviation is ≤2ms and the SCADA data clock deviation is ≤200ms.

[0118] Missing rate checks can be performed to determine whether a missing rate of ≤1%-5% for a single class of data is valid.

[0119] The value range check can be performed because the value range of the 220kV bus voltage is 198-242kV. The check can be performed to determine whether the 220kV bus voltage in the original data falls within this range.

[0120] Over-limit checks can be performed on a 110kV line where the thermal stability limit of the current is 1200A. The checks determine whether the current in the original data exceeds this upper limit.

[0121] Dimensional consistency checks can include: unifying power units to MW and temperature units to ℃, avoiding the mixing of ℉ and ℃.

[0122] It should be noted that before collecting multi-source heterogeneous data, the maximum allowable clock deviation threshold and missing rate threshold need to be set to fixed configurations; at the same time, each time slice needs to perform time consistency, missing rate, value range, out-of-limit and unit consistency checks; any record that fails any of these checks will be directly removed and marked with a data quality label and will not proceed to subsequent processing.

[0123] For example, a power grid dispatch center deploys this method, and the specific operations are as follows:

[0124] Raw data was continuously collected from 150 PMU devices, 300 SCADA measuring points, 50 micro-weather stations, and the dispatching system. All device clocks were calibrated to the dispatch center's master clock via the PTP protocol, ensuring PMU clock deviation ≤2ms and SCADA clock deviation ≤200ms. Data matching was performed with one second corresponding to one time slice. In one time slice, one of the three SCADA line power data was missing, a missing rate of 33%, and the SCADA power data for that time slice was removed. In another time slice, the PMU voltage phase angle unit was incorrectly written as degrees and minutes; this error was directly isolated and labeled. Finally, a usable dataset of 30 consecutive minutes was obtained, containing qualified multi-source data for each one-second time slice, with no time misalignments or excessive missing data.

[0125] S102. Verify the operating status of the target power grid based on the available dataset. When the verification result of the operating status of the target power grid indicates that the target power grid has undergone a change in operating condition, extract the first real-time qualified data after the change in operating condition from the available dataset.

[0126] In this step, the target power grid is identified as having undergone a change in operating conditions based on the available dataset. After confirming that a change in operating conditions has occurred, the real-time qualified data after the change is extracted to provide data raw materials for the subsequent construction of the shadow benchmark, thereby ensuring that the shadow benchmark is compatible with the new operating conditions.

[0127] Alternatively, one possible implementation method for determining the operating condition switch and acquiring the first real-time qualified data is as follows:

[0128] S1021. Based on the available dataset, obtain the second real-time qualified data for multiple consecutive sampling periods.

[0129] In this step, qualified data from the most recent consecutive sampling periods are extracted from the available dataset to form second real-time qualified data, which is used for condition switching judgment. The most recent consecutive sampling periods refer to multiple consecutive sampling periods including the current time. For example, if the current time is 0, and the duration of each sampling period is 5 seconds, and a total of 3 consecutive sampling periods are collected, then the sampling periods are: -14 to -10, -9 to -5, and -4 to 0.

[0130] S1022. When the second real-time qualified data indicates that the status of the circuit breaker in the target power grid has changed, or when the rate of change of the injected power of the power node in the target power grid reaches a preset rate of change threshold, it is determined that the operating condition of the target power grid has changed.

[0131] In this step, the second real-time qualified data collected during the operating condition switching determination must cover at least three consecutive sampling periods to avoid erroneous judgments caused by instantaneous fluctuations. An operating condition switching can be determined to have occurred in the target power grid if either of the following two conditions is met:

[0132] The second set of real-time qualified data shows that the circuit breaker status changes continuously for multiple cycles.

[0133] The absolute rate of change of the injected power at a power node within the determination window exceeds a preset power change rate threshold. A power node refers to a line node or equipment node in the target power grid. The determination window refers to a time window corresponding to multiple consecutive sampling periods. For example, if the preset power change rate threshold is set to 5%, and it is determined based on the second real-time qualified data that there are power nodes in the target power grid whose absolute rate of change exceeds 5% over three sampling periods, then it is determined that a change in operating condition has occurred in the target power grid.

[0134] S1023. Based on the available dataset, obtain the first real-time qualified data after the operating condition switch.

[0135] In this step, when it is determined that there is a change in operating conditions in the target power grid, all qualified data after the time corresponding to the change in operating conditions are extracted from the available dataset and used as the first real-time qualified data for the subsequent construction of the shadow benchmark.

[0136] For example, in the operation of a regional power grid, the lower limit of the continuous sampling period is set to 3, and the duration of the sampling period is set to 1 second; the preset power change threshold is set to 5%. The second set of qualified real-time data for the most recent 3 periods are extracted, namely 10:00:00, 10:00:01, and 10:00:02. It is found that the circuit breaker status of a certain 220kV line changes from closed at 10:00:00 to open at 10:00:01, and remains open at 10:00:02, satisfying the conditions for a topology change, and a condition switch is determined to have occurred. All qualified data after 10:00:02 are extracted to form the first set of qualified real-time data, covering 10 minutes of data after the switch.

[0137] S103. Based on the first real-time qualified data after the operating condition switch, construct the shadow benchmark of the target power grid, and use the accumulated third real-time qualified data after the operating condition switch in the available data to perform a replay verification on the shadow benchmark and generate the replay verification result.

[0138] In this step, the shadow baseline is a candidate baseline adapted to the new operating condition, constructed based on the first real-time qualified data after the operating condition switch. Replay verification verifies the reliability of the shadow baseline using accumulated data after the switch, ensuring that the shadow baseline can accurately identify abnormal and normal data under the new operating condition, thus avoiding misjudgments after the baseline switch.

[0139] Alternatively, one possible implementation of constructing a shadow benchmark and performing replay verification to obtain the replay verification result is as follows:

[0140] S1031. Based on a preset fixed-length sliding window and a preset fixed step size, statistical modeling is performed on the first real-time qualified data to obtain a shadow benchmark.

[0141] In this step, the specific methods for statistical modeling can be as follows:

[0142] a1. Based on a preset fixed-length sliding window and a preset fixed step size, the first real-time qualified data is sliced ​​to obtain data slices corresponding to multiple sliding windows.

[0143] a2. Use a time series statistical model to fit the normal distribution pattern of power grid operating parameters and cross-coupling relationship under the new operating conditions for the data in each data slice.

[0144] In this step, the time-series statistical model used can be an ARIMA model or a Gaussian mixture model (GMM). The normal distribution pattern of power grid operating parameters can be a parameter distribution pattern such as the normal range of bus power. The cross-coupling relationship can be a relationship such as the correlation strength between wind speed and line power.

[0145] a3. Based on the normal distribution pattern, cross-coupling relationship, and physical constraint boundary of the target power grid, a shadow benchmark is constructed.

[0146] S1032. Extract the third real-time qualified data accumulated after the working condition switch from the available dataset.

[0147] In this step, the accumulated third real-time qualified data refers to the data accumulated in the available dataset after the operating condition switch, and the time span of the data in the third real-time qualified data must be greater than the time span of the first real-time qualified data.

[0148] For example, the current time is 11:00:00; the time of the working condition switch is 10:00:00; assuming the time of shadow baseline construction is 10:30:00; and the time of shadow baseline replay verification is 11:00:00; then the first real-time qualified data refers to the qualified data in the available dataset from 10:00:00 to 10:30:00; the third real-time qualified data refers to the qualified data in the available dataset from 10:00:00 to 11:00:00. The time span of the third real-time qualified data must be longer than that of the first real-time qualified data, so as to avoid the self-verification caused by using the first real-time qualified data for replay verification, which would fail to detect the misjudgment problem of the shadow baseline in subsequent new scenarios.

[0149] It should be noted that the third real-time qualified data can also be qualified data from the moment the emergency baseline was constructed to the present moment.

[0150] S1033. Input the third real-time qualified data into the shadow benchmark, and use the shadow benchmark to determine whether multiple multi-source heterogeneous data in the third real-time qualified data belong to the normal operating state, and obtain multiple sets of discrimination results.

[0151] In this step, the third real-time qualified data includes multi-source heterogeneous data from multiple time slices. For the multi-source heterogeneous data of each time slice, a shadow benchmark is used for verification. The shadow benchmark determines whether the data belongs to the normal operating state or the abnormal state on a time slice-by-time according to its own fitted normal law, and obtains multiple sets of discrimination results.

[0152] S1034. Calculate the false alarm rate and false negative rate of the shadow benchmark based on the discrimination results, and generate the playback verification results.

[0153] In this step, the false alarm rate and false negative rate are calculated based on the discrimination results and the true state of the data. The true state of the data refers to the historical annotations corresponding to the multi-source heterogeneous data of each time slice, which can be normal data and abnormal data.

[0154] Optionally, the false alarm rate equals the number of times normal data is judged as abnormal, divided by the total number of times normal data occurs. The false negative rate equals the number of times abnormal data is judged as normal, divided by the total number of times abnormal data occurs. The combination of the false alarm rate and the false negative rate yields the playback verification result of the shadow benchmark.

[0155] For example, a sliding window length of 2 hours and a step size of 1 minute are set. A GMM model is used to model the first real-time qualified data to obtain a shadow baseline: under this operating condition, the normal power range of adjacent buses is 80-120MW, and the correlation coefficient between wind speed and line power is 0.75. The third real-time qualified data accumulated for 1 hour after the switchover is extracted, totaling 3600 time slices, of which 3580 are true normal data and 20 are true abnormal data. The 3600 data are input into the shadow baseline, and the judgment results are: 6 out of 3580 normal data are falsely judged as abnormal, and 1 out of 20 abnormal data is missed. The false alarm rate is calculated as 6 / 3580×100%≈0.17%, and the missed detection rate is 1 / 20×100%=5%. Based on the calculated missed detection rate and false alarm rate, a playback verification result is generated.

[0156] S104. If the playback verification result meets the preset switching conditions, the operating reference currently used by the target power grid will be updated to the shadow reference.

[0157] In this step, the operating baseline refers to the reference standard adapted to the target power grid's operating conditions. When the shadow baseline passes playback verification, the current baseline used by the power grid is replaced with the shadow baseline, thereby achieving synchronous adaptation between the baseline and operating conditions and avoiding monitoring blind spots in data auditing. The operating conditions of the target power grid refer to the stable and reproducible normal operating state of the target power grid under defined boundary conditions. Defined boundary conditions refer to the fixed key parameters of the power grid, which can include load level parameters, power output distribution parameters, network topology, and the commissioning / discontinuation status of power grid equipment. When the defined boundary conditions of the target power grid change, the operating conditions of the target power grid can be determined to have changed based on its operating status. The operating baseline corresponding to the operating conditions before the change and the operating conditions after the change may not be compatible; therefore, the baseline needs to be updated when the shadow baseline meets preset switching conditions to ensure the matching degree between the operating baseline and the operating conditions.

[0158] Optionally, the preset switching conditions include: the false alarm rate in the playback verification results does not exceed the preset false alarm rate, and the missed detection rate in the playback verification results does not exceed the preset missed detection rate. The preset false alarm rate can be a set upper limit for the false alarm rate, and the missed detection rate can be the missed detection rate of the target power grid's current operating benchmark, or it can be a set upper limit for the missed detection rate.

[0159] In this step, the construction of the target power grid benchmark can also be managed in buckets based on environmental information, with each bucket being maintained independently. This makes the benchmark more suitable for the current environment and avoids situations where the benchmark becomes unsuitable due to environmental changes.

[0160] For example, the bucket management method can be as follows: the construction of the current operating benchmark used by the power grid refers to the current benchmark reference established based on a two-hour sliding window and a one-minute step size within the stable operating range; at the same time, the benchmark is divided into buckets according to season, day and night, weather level and load segment, and each bucket maintains the benchmark distribution independently; if the sample size of a bucket is less than 500, the bucket is suspended from updating and the current version of the benchmark is retained.

[0161] Optionally, after switching the current benchmark to the shadow benchmark, real-time qualified data can be collected for a period of time after the shadow benchmark switch. This data can be used to further replay and verify the operation of the shadow benchmark. When the false alarm rate and false false alarm rate of the shadow benchmark increase and exceed the preset false alarm rate and preset false alarm rate, a version rollback is required, switching the current benchmark from the shadow benchmark to the previous version of the benchmark. Therefore, when replacing the benchmark, the historical benchmark needs to be version-numbered and stored in a preset database for easy subsequent version rollback.

[0162] The reference maintenance method for power grid operation state switching provided in this application continuously collects multi-source heterogeneous data of the target power grid within the current time period, performs unified time standard and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset. Based on the usable dataset, it detects whether the target power grid has undergone an operation state switch. When an operation state switch is determined to have occurred, it extracts the first real-time qualified data after the operation state switch from the usable dataset and uses the first real-time qualified data to construct a shadow reference for the target power grid. When it is detected that the playback verification result of the shadow reference meets the preset switching conditions, the reference currently used by the target power grid is replaced with the shadow reference. Compared with the prior art, when it is determined that the operation state of the target power grid has switched, this application first constructs a shadow reference that meets the operation state after the switch based on continuously collected data, while maintaining the current reference; when it is confirmed that the shadow reference meets the switching conditions, the current reference is switched to the shadow reference. There is no need to wait for threshold hysteresis adjustment, and it can continuously maintain the audit of multi-source heterogeneous data, thereby avoiding the technical problem of low accuracy of risk audit caused by the inability to audit some data, and achieving the technical effect of improving the accuracy of risk audit.

[0163] Figure 2 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 2 Based on the above embodiments, this application describes how to construct the coupling diagram of the target power grid, such as... Figure 2 As shown, the method includes:

[0164] S201. Extract core elements from the available datasets, including power grid core data, meteorological data, and planning data, and construct a cross-domain heterogeneous graph.

[0165] In this step, the cross-domain heterogeneous graph is a graph structure that integrates multiple elements from the power grid core domain, meteorological domain, and planning domain. By defining nodes and edges, it intuitively presents the cross-domain coupling relationships, providing a basic structure for subsequent physical constraint pruning and risk index calculation. Here, nodes refer to multiple types of elements, and edges refer to the relationships between elements.

[0166] Alternatively, one possible implementation of constructing a cross-domain heterogeneous graph is as follows:

[0167] b1. Define the core domain nodes of the power grid, the meteorological domain nodes, and the planning domain nodes.

[0168] In this step, the core domain nodes of the power grid include: transmission lines, substations, busbars, tie lines, and other electrical equipment of the power grid. The meteorological domain nodes include: wind speed, wind direction, ambient temperature, icing thickness, lightning density, and solar irradiance. The planning domain nodes include: unit output plans, AGC setpoints, planned power flow, and planned input / output of tie lines.

[0169] For example, the power grid domain nodes can be: 220kV line L1, 500kV bus M2, and the tie line between area A and area B; the meteorological domain nodes can be: wind speed V1 in the line corridor and temperature T1 around the substation; the planning domain nodes can be: planned output of unit P1 and planned input and output of tie line L1.

[0170] b2. Construct edges between nodes.

[0171] In this step, the edge construction method is as follows: based on the data extracted from the available dataset, multiple fixed time lag intervals are obtained, such as 0-1 hour and 1-2 hour time lag intervals. Within each time lag interval, the relevance loudness and causal strength of the data in different domain nodes are calculated, and the comprehensive association strength is synthesized according to preset weights.

[0172] b3. Retain edges whose overall association strength is greater than or equal to the preset statistical significance level, and whose number of edges connected to a single node is less than or equal to the preset upper limit.

[0173] In this step, the preset statistical significance level can be set to 0.05, and the preset upper limit can be set to 10. The purpose of setting the preset upper limit is to avoid node redundancy, and the purpose of setting the preset statistical significance level is to identify edges with low correlation and reduce the complexity of cross-domain heterogeneous graphs.

[0174] S202. Based on the preset core physical rules, perform pruning on the cross-domain heterogeneous graph and delete the edges that violate the physical constraints.

[0175] In this step, based on the preset core physical rules of the target power grid, we propose connections in the cross-domain heterogeneous graph that are statistically related but physically infeasible, to ensure that the coupled graph conforms to the actual operating rules of the target power grid and avoid the occurrence of physical contradictions.

[0176] Alternatively, one possible implementation for deleting edges that violate physical constraints is as follows:

[0177] c1. Determine the preset core physical rules based on the target power grid.

[0178] In this step, the preset core physical rules include: AC power flow balance constraints, line thermal stability limit constraints, bus voltage permissible range constraints, and bus phase angle compatibility constraints.

[0179] For example, the AC power flow balance constraint can be: the sum of the inflow active / reactive power of any bus = the sum of the outflow active / reactive power.

[0180] The thermal stability limit constraint for transmission lines can be: the current / power of transmission lines and tie lines ≤ the thermal stability limit value of the equipment.

[0181] The allowable range constraint for bus voltage can be: the bus voltage value is within a preset range.

[0182] The bus phase angle compatibility constraint can be: the voltage phase angle difference between adjacent buses ≤ the safety threshold of the corresponding voltage level. For example, the safety threshold of a 220kV power grid can be set to 20°.

[0183] c2. For the edges of the cross-domain heterogeneous graph, simulate the corresponding coupling relationship, calculate the operating parameters of the target power grid under the relationship, and delete the edge if any of the preset core physical rules are violated, and mark the future constraint category of the edge.

[0184] S203. The pruned cross-domain heterogeneous graph is determined as the coupling graph of the target power grid.

[0185] In this step, the pruned cross-domain heterogeneous graph is determined as a physically feasible coupled graph. This graph not only retains the real coupling relationship of multi-domain elements, but also conforms to the laws of electrical physics, and is used for the calculation of subsequent risk quantification indicators.

[0186] Optionally, one possible implementation of determining the coupling graph is to name the pruned cross-domain heterogeneous graph the target power grid coupling graph and store its node information and edge information.

[0187] Figure 3 A flowchart illustrating the benchmark maintenance method for switching power grid operating states provided in this application. Figure 3 Based on the above embodiments, this application describes how to generate audit scores for a target power grid, such as... Figure 3 As shown, the method includes

[0188] S301. Extract the fourth real-time qualified data after the baseline switch is completed from the available dataset.

[0189] In this step, the fourth real-time qualified data refers to the latest real-time qualified data in the available dataset after the benchmark switch is completed. It is used as the data source for calculating risk quantification indicators to ensure that the indicators can reflect the real-time operating status of the power grid after the benchmark switch.

[0190] It should be noted that when it is determined that the shadow reference does not meet the preset switching conditions, the current reference of the target power grid will still be used, and the fourth real-time qualified data will still be the latest real-time qualified data.

[0191] Optionally, the fourth real-time qualified data can be obtained by extracting qualified data of a preset time length from the available dataset in real time, ensuring that the data can cover the elements of the power grid core domain, meteorological domain, and planning domain.

[0192] For example, the current time is 11:00:00; the reference switchover completion time is 10:40:00; after the reference switchover is completed, qualified data within 10 minutes containing the current time is extracted to form the fourth real-time qualified data. The fourth real-time qualified data refers to the qualified data from 10:50:00 to 11:00:00. The data in the available dataset is continuously updated over time, and the latest qualified data in the available dataset corresponds to the multi-source heterogeneous data of the target power grid at the current time, thus ensuring that the fourth real-time qualified data extracted from the available dataset includes the latest data at the current time.

[0193] S302. Based on the coupling diagram of the target power grid and the fourth real-time qualified data, the index values ​​of three sets of risk quantification indicators are calculated.

[0194] In this step, the risk quantification indicators refer to the low-frequency energy increment, feasibility gap, and the credibility of the data source.

[0195] Optionally, the calculation method for the risk quantification indicator can be:

[0196] d1. Calculate the low-frequency energy increment of the spectrum, specifically:

[0197] Based on the node and edge structure of the coupled graph, construct the graph Laplacian matrix.

[0198] The fourth real-time qualified data is input into the graph Laplacian matrix to calculate the low-frequency energy of the graph Laplacian matrix at the current time.

[0199] The difference between the calculated low-frequency energy and the reference low-frequency energy of the target power grid is calculated to obtain the spectral low-frequency energy increment.

[0200] d2. Calculate the feasibility gap, specifically: without changing the measurements, introduce the minimum total slack to balance the power flow of the target power grid and satisfy the security constraints of the target power grid. The optimal target value at this time is the feasibility gap at the current moment.

[0201] In this step, the feasibility gap is calculated as follows:

[0202] Minimizing the total relaxation amount is taken as the objective optimization function, and the constraints are set as power flow balance and safety constraints. The objective optimization function is shown in Equation 1:

[0203]

[0204] Where, min x,s This refers to finding the minimum value of variables x and s; x refers to the decision variables representing the real-time operating parameters of the power grid; s refers to the slack quantity. TS refers to the 1-norm of the slack, used to minimize the slack and ensure minimal adjustment. F(x) refers to the power flow balance function, used to constrain the AC power flow balance of the power grid; G(x) refers to the safety constraint function, which can be line thermal stability constraint, bus voltage constraint, or phase angle compatibility constraint. G(x)≤s means that if variable x does not meet these constraints, the slack s is needed to compensate for the gap and make the constraints valid. s≥0 indicates that the slack is a non-negative constraint.

[0205] Substituting the fourth real-time qualified data into Formula 1, the minimum total relaxation amount is obtained, which is the feasibility gap.

[0206] d3. Calculate the reliability of the data source, specifically:

[0207] Extract quality parameters from the fourth real-time qualified data, such as timestamp deviation, communication latency, most recent calibration time, and device self-test results.

[0208] The quality parameters are scored according to the preset mapping rules.

[0209] The scores for the quality parameters are weighted and averaged, and then mapped to a range of 0 to 1 to obtain a weighted score, which represents the credibility of the data source.

[0210] S303. Based on the benchmark currently used by the target power grid, obtain the benchmark distribution corresponding to the benchmark.

[0211] In this step, the construction of the benchmark distribution refers to calculating a large number of samples of three types of indicators using early historical qualified data after the change in operating conditions during the training phase of the shadow benchmark. The samples are then divided into four-dimensional buckets according to season, day / night time, meteorological level, and load segment, and the benchmark distribution is obtained using kernel density estimation or parameter fitting. The benchmark distribution can be obtained by directly extracting it from the database corresponding to the benchmark.

[0212] S304. Based on the indicator values ​​of the three sets of risk quantification indicators, generate the joint distribution of the indicators corresponding to the indicator values.

[0213] In this step, the joint distribution of indicators refers to the multidimensional statistical distribution of the three sets of risk quantification indicators corresponding to the fourth real-time qualified data. It is used to reflect the combined characteristics of the three types of indicators under the current real-time state and to compare with the benchmark distribution.

[0214] Optionally, the joint distribution of indicators can be generated by: extracting multiple sets of three-category indicator samples from the fourth real-time qualified data; using the kernel density estimation method to perform multidimensional fitting on the multiple sets of three-category indicator samples to obtain the joint probability density function of the three-category indicators, i.e., the joint distribution of indicators.

[0215] S305. The distribution deviation value is calculated based on the joint distribution of indicators and the benchmark distribution.

[0216] In this step, the distribution deviation value refers to the quantitative difference between the joint distribution of indicators and the baseline distribution. It is used to reflect the degree of deviation between the current indicator distribution and the safety distribution of the new operating condition. The larger the distribution deviation value, the higher the risk represented by the current target power grid data.

[0217] Optionally, the distribution deviation can be calculated by calculating the Wasserstein distance or Jensen-Shannon distance between the joint distribution of the indices and the baseline distribution, thus obtaining the corresponding deviation value. The closer the Jensen-Shannon distance is to 1, the greater the distribution difference; the larger the Wasserstein distance, the greater the difference.

[0218] S306. Based on the distribution deviation value and the index values ​​of three sets of risk quantification indicators, the audit risk score of the target power grid is calculated.

[0219] In this step, the process of calculating the audit risk score is as follows: The index values ​​of the three sets of risk quantification indicators are standardized and mapped to a range of 0 to 1. Weighting coefficients for calculating the audit risk score are determined using historical risk audit data of the target power grid. The audit risk score is then calculated based on the distribution deviation, the standardized index values, and the weighting coefficients.

[0220] Optionally, the audit risk score can be calculated as shown in Formula 2:

[0221]

[0222] Where R(t) refers to the audit risk score at time t; W1, W2, and W3 refer to the weight coefficients of each risk dimension, where W1+W2+W3=1; used to reflect the importance of each risk dimension. This refers to the degree of distribution deviation; This refers to the low-frequency energy increment of the spectrum; C(t) refers to the feasibility gap; C(t) refers to the credibility of the data source.

[0223] Based on the above embodiments, this application also provides a method for power grid risk auditing. Figure 4 A flowchart illustrating the power grid risk auditing method provided in this application embodiment is shown below. Figure 4 As shown, the method includes:

[0224] A1. Obtain historical operational data for risk assessment corresponding to the target power grid.

[0225] In this step, historical operational data refers to multi-source heterogeneous data over a period of time that includes real-time data at the current moment. Historical operational data can be based on the above. Figure 1The available dataset construction methods shown are used to construct the dataset, thereby obtaining historical running data suitable for subsequent data processing.

[0226] A2. Divide the historical operation data into multiple non-overlapping evaluation blocks according to the timeline.

[0227] In this step, the time intervals between multiple evaluation blocks are continuous and non-overlapping, and the duration of each evaluation block can be set to 10 minutes. Each evaluation block contains multiple time slices, and the length of each time slice can be set to 1 second.

[0228] A3. Calculate the average of the audit risk scores for multiple time slices within each assessment block to obtain the block average risk score for each assessment block.

[0229] In this step, the audit risk score for the time slice can be calculated using the above method. Figure 3 The method shown can be specifically calculated using Formula 2 above.

[0230] For example, if there are 600 time slices in an evaluation block, the block average risk score of the evaluation block is the average of the audit risk scores of the 600 time slices.

[0231] A4. Based on the time sequence of multiple assessment blocks, determine whether the average risk score of each block exceeds the alarm threshold.

[0232] In this step, the alarm threshold is determined based on the error rate constraint of the target power grid. The error rate includes Type I error rate and Type II error rate. Type I error rate corresponds to the false alarm rate, and Type II error rate corresponds to the missed detection rate.

[0233] For example, the number of evaluation blocks is 1000, and the Class I error rate is set to 0.01, meaning that a maximum of 10 false positives are allowed in 1000 security scenarios. The scores of the 100 evaluation blocks are mainly distributed between 0-2 and 0.5, conforming to a normal distribution N(0.35, 0.05²). By looking up the normal distribution table, the alarm threshold corresponding to the Class I error rate is found to be 0.46, thus determining the alarm threshold.

[0234] Similarly, assuming there are 500 evaluation blocks and a Type II error rate of 0.05, this means that at most 25 missed detections are allowed in 500 abnormal scenarios. The score distribution of these 500 evaluation blocks is N(0.65, 0.08²), and by looking up a table, we can determine that the alarm threshold cannot be higher than 0.52.

[0235] Combining the alarm thresholds of these two categories, the final alarm threshold can be the average of the two, therefore the alarm threshold is set to 0.5.

[0236] A5. When the number of evaluation blocks that continuously exceed the alarm threshold reaches a preset number, it is determined that there is a potential anomaly in the target power grid; if only the average risk score of a single evaluation block exceeds the alarm threshold, the subsequent process will not be executed.

[0237] In this step, the pre-set quantity is determined based on the error rate constraint of the target power grid.

[0238] For example, consider a Class I error rate of 0.01, a Class II error rate of 0.05, and an alarm threshold of 0.5. If only one evaluation block is required to exceed the alarm threshold to determine an anomaly, the Class I error rate is satisfied, but there may be false alarms due to a single evaluation block exceeding the alarm threshold caused by noise. If two consecutive evaluation blocks are required to exceed the alarm threshold before determining an anomaly, the Class I error rate is satisfied, but the Class II error rate increases. Therefore, it is necessary to find the optimal alarm threshold and the pre-set number while satisfying both the Class I and Class II error rates. This ensures that the error rate constraints are met without causing alarm delays due to an excessive number of consecutive blocks.

[0239] A6. When a potential anomaly is determined in the target power grid, perform a cross-domain invariance test and obtain the test results.

[0240] Alternatively, the method for cross-domain invariance testing is as follows:

[0241] The weights of the connections between meteorological domain nodes and electrical core domain nodes are extracted from the coupled graph of the target power grid. The rate of change of the current connection weight compared to the weights of connections at historical time points is calculated to obtain the edge weight change rate. If the edge weight change rate is less than the edge weight change rate threshold for the meteorological domain, it is determined that the potential anomaly of the target power grid is not caused by meteorological disturbances. The edge weight change rate threshold is a preset parameter set based on historical operating data.

[0242] The weights of the edges connecting the planning domain nodes and the electrical core domain nodes are extracted from the coupling graph of the target power grid. The rate of change of the current edge weight relative to the edge weights in historical operating data is calculated to obtain the edge weight change rate corresponding to the planning domain. When the edge weight change rate is lower than the edge weight change rate threshold of the planning domain, it is determined that the potential anomaly of the target power grid is not caused by planned disturbances. The edge weight change rate threshold of the planning domain is a parameter pre-set based on historical operating data.

[0243] Real-time multi-source heterogeneous data of the target power grid is acquired, and key parameters of the electrical core domain are extracted. Based on these key parameters, it is determined whether the line current exceeds the thermal stability limit, whether the bus voltage exceeds the allowable bus voltage range, and whether the line power flow is balanced. If all conditions are met, it is determined that the potential anomaly of the target power grid is not caused by the constraint tension of external disturbances.

[0244] A7. When the test results indicate that the test has passed, identify the key sources of influence that caused the target power grid anomaly.

[0245] In this step, the key impact sources are identified as follows: multiple key objects in the target power grid are identified, which can be PMU devices, lines, weather stations, and planned data nodes. The impact of removing each key object is simulated one by one, and the difference between the average risk score of the assessment block before removal and the average risk score after removal is calculated. The larger the difference, the greater the marginal contribution of the key object, indicating that the object has a greater impact on the anomaly.

[0246] Based on the marginal contribution of each key object, sort them from highest to lowest, and select N key objects as key influence sources. Here, N is a positive integer greater than 0. N can be set to 10.

[0247] A8. Repair the key impact sources and record the repair results.

[0248] In this step, each key source of impact is repaired in order of decreasing impact and decreasing reversibility. After each repair, a replay verification is performed on the repaired baseline. When the replay verification detects that the false alarm rate or false negative rate of the baseline has deteriorated, a version rollback is performed on the baseline.

[0249] The remediation of key impact sources is divided into four stages: state estimation replacement, timescale realignment, cross-source replacement, and historical distribution backfilling. Remediation refers to fixing errors at the data level to avoid misjudgments caused by inaccurate data detection.

[0250] Optionally, state estimation replacement refers to replacing the locatable measurement offset with a weighted least squares state estimation result. For example, the voltage phase angle offset of the PMU can be replaced with its corresponding state estimate. The state estimate is calculated based on historical qualified data.

[0251] Time stamp realignment refers to resampling or phase alignment of data with detected time stamp misalignment. For example, if the power data of line L5 deviates from the timestamp of other data by 50ms, resampling or phase alignment is required.

[0252] Cross-source replacement refers to replacing faulty data based on the physical mapping of redundant measurement points or adjacent measurements. For example, a bus has two PMUs monitoring voltage; the faulty PMU data is replaced with data from the normal PMU. Line L5 has no redundant measurement points; the power data from the adjacent line L4 is used to estimate the power of L5 based on the line impedance relationship.

[0253] Historical distribution backfilling refers to conditional backfilling based on the baseline distribution of the current working condition buckets if the first three levels of repair have all failed. The failure of the first three levels of repair means that the data of the key impact sources has no redundant measurement points and the physical mapping is unreliable.

[0254] For example, historical distribution backfilling can be performed by: finding the bucket corresponding to the current operating condition; randomly selecting samples that conform to the baseline distribution; and backfilling the missing or faulty data in the key data source.

[0255] Figure 5 A schematic diagram of the structure of the reference maintenance device for switching power grid operating states provided in this application is shown below. Figure 5 As shown, the reference maintenance device for switching power grid operating states provided in this embodiment includes:

[0256] The acquisition module 501 is used to continuously collect multi-source heterogeneous data of the target power grid in the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset.

[0257] The first processing module 502 is used to verify the operating status of the target power grid based on the available dataset. When the verification result of the operating status of the target power grid indicates that the target power grid has undergone a change in operating condition, the first real-time qualified data after the change in operating condition is extracted from the available dataset.

[0258] The second processing module 503 is used to construct a shadow benchmark of the target power grid based on the first real-time qualified data after the operating condition switch, and to perform a replay verification on the shadow benchmark using the accumulated third real-time qualified data after the operating condition switch in the available data set, and generate a replay verification result.

[0259] The third processing module 504 is used to update the operating reference currently used by the target power grid to the shadow reference when the playback verification result meets the preset switching conditions.

[0260] Alternatively, in one possible implementation, the acquisition module 501 is further configured to:

[0261] Continuously collect multi-source heterogeneous data of the target power grid during the current period; among which, multi-source heterogeneous data includes core power grid data, meteorological data, and planning data.

[0262] Based on multiple data sources corresponding to multi-source heterogeneous data, a unified time base is established for multiple data sources.

[0263] The raw data from multiple data sources are matched according to time slices to obtain multi-source heterogeneous data under different time slices.

[0264] For each time slice, a quality check is performed on the multi-source heterogeneous data. The multi-source heterogeneous data that passes the check is retained and combined according to the time order of the time slice to obtain a usable dataset.

[0265] Optionally, in one possible implementation, the first processing module 502 is further configured to:

[0266] Based on the available dataset, second real-time qualified data is obtained for multiple consecutive sampling periods.

[0267] When the second real-time qualified data indicates a change in the status of a circuit breaker in the target power grid, or when the rate of change of the injected power of a power node in the target power grid reaches a preset rate of change threshold, it is determined that the operating condition of the target power grid has switched.

[0268] Based on the available dataset, obtain the first real-time qualified data after the operating condition switch.

[0269] Alternatively, in one possible implementation, the second processing module 503 is further configured to:

[0270] Based on a preset fixed-length sliding window and a preset fixed step size, statistical modeling is performed on the first real-time qualified data to obtain a shadow benchmark.

[0271] Optionally, in one possible implementation, the shadow baseline is replayed and verified using the accumulated third real-time qualified data after the change of operating conditions in the available dataset, and the replay verification result is generated, including:

[0272] Extract the third real-time qualified data accumulated after the change of operating conditions from the available dataset.

[0273] The third real-time qualified data is input into the shadow benchmark. The shadow benchmark is used to determine whether multiple multi-source heterogeneous data in the third real-time qualified data belong to the normal operating state, and multiple sets of discrimination results are obtained.

[0274] The false alarm rate and false negative rate of the shadow benchmark are calculated based on the discrimination results, and the playback verification results are generated.

[0275] Optionally, in one possible implementation, preset switching conditions include:

[0276] The false alarm rate in the playback verification results did not exceed the preset false alarm rate, and the false detection rate in the playback verification results did not exceed the preset false detection rate.

[0277] Alternatively, in one possible implementation, the acquisition module 501 is further configured to:

[0278] Core elements from power grid core data, meteorological data, and planning data are extracted from available datasets to construct a cross-domain heterogeneous graph.

[0279] Based on preset core physical rules, pruning is performed on cross-domain heterogeneous graphs to delete edges that violate physical constraints.

[0280] The pruned cross-domain heterogeneous graph is used as the coupling graph of the target power grid.

[0281] The coupling graph is used for the subsequent calculation of risk quantification indicators.

[0282] Alternatively, in one possible implementation, the third processing module 504 is further configured to:

[0283] Extract the fourth real-time qualified data after the benchmark switch is completed from the available dataset.

[0284] Based on the coupling diagram of the target power grid and the fourth set of qualified real-time data, the index values ​​of three sets of risk quantification indicators are calculated. The risk quantification indicators refer to the low-frequency energy increment, feasibility gap, and data source credibility.

[0285] Based on the benchmark currently used by the target power grid, obtain the benchmark distribution corresponding to the benchmark.

[0286] Based on the indicator values ​​of the three sets of risk quantification indicators, a joint distribution of the indicators corresponding to the indicator values ​​is generated.

[0287] The distribution deviation value is calculated based on the joint distribution of the indicators and the benchmark distribution.

[0288] Based on the distribution deviation value and the index values ​​of three sets of risk quantification indicators, the audit risk score of the target power grid is calculated.

[0289] The apparatus provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0290] Figure 6 A schematic diagram of the structure of the electronic device provided in this application. Figure 6 As shown, the electronic device provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the device further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0291] In the specific implementation process, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to execute the aforementioned benchmark maintenance method or method for switching power grid operating states.

[0292] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0293] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0294] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0295] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0296] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0297] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0298] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0299] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0300] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0301] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0302] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0303] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0304] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0305] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A benchmark maintenance method for switching power grid operating states, characterized in that, include: Continuously collect multi-source heterogeneous data of the target power grid within the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset; The target power grid is verified based on the available dataset. When the verification result of the target power grid's operating status indicates that the target power grid has undergone a change in operating condition, the first real-time qualified data after the change in operating condition is extracted from the available dataset. Based on the first real-time qualified data after the operating condition switch, a shadow benchmark of the target power grid is constructed, and the shadow benchmark is replay-verified using the accumulated third real-time qualified data after the operating condition switch in the available data set, generating a replay-verification result. If the playback verification result meets the preset switching conditions, the operating reference currently used by the target power grid will be updated to the shadow reference; wherein, the operating reference refers to a reference standard adapted to the operating conditions of the target power grid.

2. The method according to claim 1, characterized in that, The process involves continuously collecting multi-source heterogeneous data from the target power grid within the current time period, and then performing unified time standard alignment and quality filtering on the multi-source heterogeneous data to obtain a usable dataset, including: Continuously collect multi-source heterogeneous data of the target power grid during the current time period; wherein, the multi-source heterogeneous data includes core power grid data, meteorological data, and planning data; Based on the multiple data sources corresponding to the multi-source heterogeneous data, a unified time base is established for the multiple data sources. The raw data from the multiple data sources are matched according to time slices to obtain multi-source heterogeneous data under different time slices; For each time slice, a quality check is performed on the multi-source heterogeneous data. The multi-source heterogeneous data that passes the quality check is retained and combined according to the time order of the time slice to obtain the usable dataset.

3. The method according to claim 1, characterized in that, The step of verifying the operating status of the target power grid based on the available dataset, and when the verification result of the operating status of the target power grid indicates that a change in operating condition has occurred, extracting the first real-time qualified data after the change in operating condition from the available dataset, includes: Based on the available dataset, obtain second real-time qualified data for multiple consecutive sampling periods; When the second real-time qualified data indicates that the status of the circuit breaker in the target power grid has changed, or when the rate of change of the injected power of the power node in the target power grid reaches a preset rate of change threshold, it is determined that the operating condition of the target power grid has changed. Based on the available dataset, obtain the first real-time qualified data after the operating condition switch.

4. The method according to claim 1, characterized in that, The construction of the shadow benchmark of the target power grid based on the first real-time qualified data after the operating condition switch includes: Based on a preset fixed-length sliding window and a preset fixed step size, statistical modeling is performed on the first real-time qualified data to obtain the shadow benchmark.

5. The method according to claim 1, characterized in that, The step of using the accumulated third real-time qualified data after the switching of operating conditions in the available dataset to perform replay verification on the shadow benchmark and generate replay verification results includes: Extract the third real-time qualified data accumulated after the operating condition switch from the available dataset; The third real-time qualified data is input into the shadow benchmark, and the shadow benchmark is used to determine whether multiple multi-source heterogeneous data in the third real-time qualified data belong to the normal operating state, and multiple sets of discrimination results are obtained. The false alarm rate and false negative rate of the shadow benchmark are calculated based on the discrimination results, and the playback verification results are generated.

6. The method according to claim 5, characterized in that, The preset switching conditions include: The false alarm rate in the playback verification results did not exceed the preset false alarm rate, and the false detection rate in the playback verification results did not exceed the preset false detection rate.

7. The method according to claim 1, characterized in that, Before performing the operational status verification of the target power grid based on the available dataset, the method further includes: Extract core elements from the available dataset, including power grid core data, meteorological data, and planning data, and construct a cross-domain heterogeneous graph. Based on preset core physical rules, the cross-domain heterogeneous graph is pruned to delete edges that violate physical constraints; The pruned cross-domain heterogeneous graph is used as the coupling graph of the target power grid; The coupling graph is used for the calculation of subsequent risk quantification indicators.

8. The method according to claim 1, characterized in that, After updating the operating reference currently used by the target power grid to the shadow reference, the method further includes: Extract the fourth real-time qualified data after the benchmark switch is completed from the available dataset; Based on the coupling diagram of the target power grid and the fourth set of qualified real-time data, the index values ​​of three sets of risk quantification indicators are calculated; the risk quantification indicators refer to the low-frequency energy increment, feasibility gap, and data source credibility. Based on the benchmark currently used by the target power grid, obtain the benchmark distribution corresponding to the benchmark; Based on the index values ​​of the three sets of risk quantification indicators, a joint distribution of the indicators corresponding to the index values ​​is generated. The distribution deviation value is calculated based on the joint distribution of the indicators and the benchmark distribution; Based on the distribution deviation value and the index values ​​of the three sets of risk quantification indicators, the audit risk score of the target power grid is calculated.

9. A reference maintenance device for switching power grid operating states, characterized in that, include: The acquisition module is used to continuously collect multi-source heterogeneous data of the target power grid in the current time period, and perform unified time standard alignment and quality screening processing on the multi-source heterogeneous data to obtain a usable dataset; The first processing module is used to verify the operating status of the target power grid based on the available dataset. When the verification result of the operating status of the target power grid indicates that the target power grid has undergone a change in operating condition, the module extracts the first real-time qualified data after the change in operating condition from the available dataset. The second processing module is used to construct a shadow benchmark of the target power grid based on the first real-time qualified data after the operating condition switch, and to perform a replay verification on the shadow benchmark using the accumulated third real-time qualified data after the operating condition switch in the available data set, and generate a replay verification result. The third processing module is used to update the operating reference currently used by the target power grid to the shadow reference when the playback verification result meets the preset switching conditions.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

Images